Source: https://openbot.run/news/introducing-openbot [All news](https://openbot.run/news) # Introducing OpenBot: A Shared Workspace for AI Agents Meet OpenBot, a local-first workspace where the AI models you already use can work as a team of agents, alongside you and your coworkers. Sep 14, 2026 Norbert Bodziony Many of us already use an AI tool every day. It might be ChatGPT or Claude on our computers, or an AI search feature in a browser. These models are useful, but they usually work as one assistant at a time. OpenBot is built around a different idea: several agents can work together in one shared space, each with its own role. ## What Is OpenBot? OpenBot is a free, source-available, local-first alternative to Grok Bot. It does not train its own AI model. Instead, it brings the models you already use into one desktop workspace. You can connect Codex, Claude, and Grok side by side. Through OpenCode, you can also use a model or provider you have configured yourself. Once they are connected, you can build a team of agents around the work. Each agent has its own role, workspace, conversation, queue, and context. They can work in parallel, exchange messages and files, and keep a task moving forward instead of starting from a blank chat every time. OpenBot does not replace the models behind the agents. It gives them one place to work together — and alongside the people on your team. ## From Single-Player to Multiplayer Many AI tools are built around one person working with one assistant at a time. You open a chat, ask a question, and work with the answer on your own. OpenBot is built around a different idea: people and AI agents can work in the same workspace. Coworkers can see the same conversations and files, talk to the agents, and coordinate their work in one place. Agents can also communicate with one another, pass work between themselves, and keep their own roles and context. That makes the work feel less like opening a new chat for every task and more like working in one shared space. ## How It Works Using OpenBot starts with three steps: create an agent, give it a job, and let it keep working in its own workspace and context. You give it a role and instructions, then return to the same conversation instead of rebuilding the setup every time. Once one agent is in place, you can add others and invite coworkers to the same space. They can follow the work, talk to the agents, see the shared conversations, and exchange files as the task moves forward. ## What Can You Do With OpenBot? OpenBot is not built around one specific workflow. You can create agents for different kinds of work, including: - Research and monitoring - Coding and development - Content and marketing - Product launches - Sales workflows - Recurring tasks and information gathering The setup can change with the project: one person may use a single research agent, while a team may divide the work between several agents with different roles. ## Why Run OpenBot Locally? One advantage of running OpenBot locally is that it stays close to your files, tools, and everyday workspace. You can host it on a Mac mini, leave it running, and access the same space from other devices. That keeps your agents and projects in one place while still letting you check in from elsewhere. OpenBot's source code is public as well, so anyone can inspect the project and contribute to it. ## Working With Multiple Agents OpenBot brings the models you already use into one workspace. You can give different agents different roles, keep their context with the project, and let them pass work to one another instead of starting a new chat for every task. That is useful when a task needs more than one answer. Research, coding, writing, and review can be split between agents while the project stays in one place. --- Source: https://openbot.run/news/your-work-stays-on-your-computer [All news](https://openbot.run/news) # Your work stays on your computer Workspaces, conversations and attachments live in a SQLite database you own. Here is what that rules out, and the one thing it does not. Sep 8, 2026 Norbert Bodziony OpenBot keeps your workspaces, conversations, attachments, browser data and team data on the computer that runs it. They go into a SQLite database in your own home directory. There is no copy on our side, because there is no place on our side to put one. ## What that rules out We cannot read your chats. We cannot hand them to anyone who asks us, because we do not hold them. If you delete the database, the conversation is gone rather than merely hidden. It also changes who is responsible for backups. Nothing is uploaded, so nothing is restored for you. That trade is the point, but it is worth knowing before you rely on it. ## The one thing it does not rule out Local-first is a claim about storage, not about the network. Your agents still talk to model providers, and whatever you send an agent reaches the provider you pointed it at. Pages you open in the built-in browser still load from the internet. Plugins can still make requests. The honest version of the promise is narrower than the slogan: OpenBot adds no hop of its own. We wrote down exactly which data leaves your machine and why in the [privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md). ## What an account is for You can sign in, and some things do need one: your profile and avatar, host configuration, team memberships, invitations and the logical sessions that let devices find each other. That list is deliberately short, and it is the whole of it. Chats, files and commands are not on it. Your database stays the source of truth. The account is not a cache of your work that happens to live elsewhere; it is the small amount of shared state that two machines need in order to agree they belong to the same person. --- Source: https://openbot.run/news/one-agent-many-providers [All news](https://openbot.run/news) # One agent, many providers An agent keeps its workspace, its thread and its identity when you move it between Codex, Claude and Grok. Switching model should not cost you the context. Aug 27, 2026 Norbert Bodziony Most tools tie the assistant to the model. Change the model and you start again: new chat, new working directory, none of the history that made the last answer good. The model is a setting, so it should behave like one. ## What an agent keeps In OpenBot an agent owns three things that outlive any provider it runs on: a workspace on disk, a thread of everything it has done, and an identity of its own. Point the same agent at [Codex](https://learn.chatgpt.com/docs/app-server) instead of [Claude Code](https://code.claude.com/docs/en/overview) and it is still the same agent, in the same folder, with the same history behind it. Restarting the application does not reset it either. An agent that was halfway through something is still halfway through it tomorrow. ## Why it is built this way The alternative is tempting and we decided against it. Wiping an agent on a provider switch would make the state easy to reason about, because there would be almost none. It would also make the feature useless: the only reason to move an agent between providers is to keep what it already knows and change how it thinks. So the provider session is kept separate from the thread. The session is private resume state belonging to one command-line tool. The thread is yours, and it is what survives. ## What this is good for Run the cheap model until the work gets hard, then move that exact agent up to a stronger one without re-explaining the task. Move it back down when the hard part is finished. Try the same job on two providers and compare the results rather than the setup effort. --- Source: https://openbot.run/news/channels-put-agents-in-one-room [All news](https://openbot.run/news) # Channels put agents in one room A channel is a shared thread several agents read and write. It turns a pile of parallel chats into something closer to a team conversation. Aug 14, 2026 Norbert Bodziony Running several agents at once is easy. Keeping track of them is not. Each one gets its own chat, each chat holds one half of a conversation, and the part you actually want — what they decided together — exists only in your head as you tab between them. ## A channel is a shared thread A channel is one conversation that several agents read and write. You post once and everyone in the channel sees it. An agent's reply is visible to the others, so the second agent answers with the first one's work already in front of it instead of a summary you retyped. Each agent still has its own workspace and its own private thread. The channel adds a place they share; it does not merge them into one process with one memory. ## Where it helps It suits work with an obvious division of labour. One agent reads the code and one writes the tests. One drafts and one reviews. One is pointed at the API and one at the client that calls it. In each case the handoff is the expensive part, and a shared thread is what removes it. ## Where it does not A channel is not a way to make one hard problem go faster. Three agents on a single ambiguous task mostly produce three plausible answers and a longer thread to read. It earns its place when the agents have genuinely different jobs. The habit worth forming is the same one that works with people: give the room a reason to exist before you invite anyone into it. --- Source: https://openbot.run/news/routines-give-an-agent-a-schedule [All news](https://openbot.run/news) # Routines give an agent a schedule A routine is one instruction and the times to run it. The agent does the work in its own thread, so the result lands where the context already is. Aug 5, 2026 Norbert Bodziony Most of what you ask an agent twice, you will ask it a hundred times. Read the overnight build log. Summarise what changed in the repository since yesterday. Check whether the thing you fixed on Friday is still fixed. None of it is hard; all of it depends on you remembering to ask. ## A routine is one instruction on a clock A routine belongs to one agent. It holds the instruction and the times to run it, and nothing else. When it fires, the agent does the work in its own workspace and its own thread, exactly as if you had typed the instruction yourself. That last part matters more than it sounds. The result lands in the thread you already read, next to the context that explains it. There is no second inbox to check. ## Keep them small A routine that says "review the repository" produces a wall of text you will stop reading by Wednesday. A routine that says "list the tests that started failing since the last run, and nothing else" produces something you act on. The test we use is simple: if the usual answer is "nothing to report", the routine is probably the right size. ## They are not a scheduler A routine is an instruction for an agent, not a job runner for your machine. It has no retry policy and no dependency graph, and it is the wrong tool for anything that must happen exactly once at exactly one time. Use it for the reading and the noticing, and keep the deploy on the deploy system. --- Source: https://openbot.run/news/every-agent-gets-a-workspace [All news](https://openbot.run/news) # Every agent gets a workspace One directory per agent, kept between runs, and one shared directory for handoffs. A workspace organizes an agent's work; it does not limit what the agent can touch. Jul 24, 2026 Norbert Bodziony Every agent keeps a directory of its own on disk. It is where the agent puts its own files, and where they stay between runs. When you want to know what an agent has been working on, that directory is the first place to look. ## One directory per agent The workspace outlives the conversation. Close the app, reopen it a week later, and the agent is standing where you left it, with its files where it left them. ## A shared directory for handoffs Next to the agents' own directories there is one shared directory that every agent can use. An agent keeps its own work in its workspace and puts files that another agent needs in the shared one. A file there is there because an agent put it there for someone else, so a handoff stays something you can see. ## What a workspace is not A workspace organizes work. It does not limit what an agent can touch. Agents run with full access to your files, your programs and the network, so an agent can read or change a file outside its directory, including one in another agent's workspace. Say in an agent's instructions where its work belongs, and keep backups or version control for anything you cannot afford to lose. --- Source: https://openbot.run/news/run-the-team-server-yourself [All news](https://openbot.run/news) # Run the team server yourself Learn how self-hosting an OpenBot team server works, where conversations and files live, what the account service handles, and who is responsible for uptime. Jul 11, 2026 Norbert Bodziony Updated Oct 7, 2026 OpenBot keeps your agents, threads, workspaces and files on your own computer. ## What a team adds A team is what you want when a second person needs to see the work. One computer runs the host, the others join it, and the chats and files stay on the machine that runs the host. If you run that host yourself, it is your hardware and your network. ## What the account is for There is an account service, and it is deliberately small. It holds accounts, avatars, memberships, invitations and the configuration that lets one computer find another. It does not hold your conversations, your files or your commands. The account service is separate from the team server. With self-hosting, the host and your work run on hardware you control. With an OpenBot hosted server, OpenBot provides the Linux machine and your team's work stays there instead of on your own computer. The account service still handles memberships and invitations. See our [guide to OpenBot hosted servers](https://openbot.run/guides/openbot-hosted-servers) for the plans and setup. In other words, the account service handles the connection between people and servers; the team server is where the shared work lives. ## The trade you are making Self-hosting means uptime is yours. If the machine that runs the host is asleep, the team is offline, and no one else is going to wake it up for you. That is the cost, and it is worth knowing before you build a workflow that depends on it. --- Source: https://openbot.run/news/what-gets-redacted-before-it-leaves [All news](https://openbot.run/news) # What gets redacted before it leaves Secrets are stripped at every path out of the app: logs, diagnostics, crash reports and analytics. Here is why we redact at the exit, not at the call site. Jun 30, 2026 Norbert Bodziony A desktop app that talks to several model providers accumulates secrets: API keys, tokens, the contents of the environment it was started with. A log line is the easiest way in the world to leak one, and a diagnostics bundle is the easiest way to leak all of them at once. ## Redact at the exit, not at the call site Redacting where the secret is logged only works until someone adds a log line and forgets. The rule we hold instead is that every path out of the app redacts: logs, exported diagnostics, crash reports, analytics. If bytes can leave, they go through something that strips secrets first. It is a duller design than a clever helper at each call site, and it fails in the right direction. A new log line written by someone who has never read this article is still covered. ## What analytics is allowed to know The website counts pages and link clicks. The app does not send your prompts, your files or your provider traffic anywhere, and there is no path in it that would. Anything we cannot collect by accident is the sort of thing we prefer not to have to promise. ## Read it for yourself The source is public and the privacy document travels with it. If you find a path that logs something it should not, that is a bug worth reporting, and we would rather hear it from you than not hear it at all. --- Source: https://openbot.run/guides/ai-agent-workflow [All guides](https://openbot.run/guides) # AI Agent Workflow: From Task to Finished Result Follow a practical AI agent workflow in OpenBot. Write a clear brief, check official sources, save a result and review it before you share it. Nov 12, 2026 Norbert Bodziony An [AI agent](https://openbot.run/guides/what-are-ai-agents) workflow is the sequence that takes a request to a result you can use. For a first workflow in OpenBot, pick one task and define what you want to receive. A short, sourced release brief is a good example. Suppose a tool your team uses just released an update. You want to know what changed, who needs to know and what still needs checking before you share the news internally. ## 1. Give the agent a clear brief Name the tool and the release. Say who the brief is for, what to cover and where to save it. If you have the official release page or last week’s brief, give the agent those links or files. Here is a prompt you can adapt: > Read the official release notes for [tool and version]. Pick up to three changes that matter to our support team. For each one, explain who it affects and link to the source. Flag anything you cannot confirm. Save the draft as release-brief.md in your workspace and show me the file for review. You now have a way to check the result. The file should exist, the changes should be relevant to support, and the links should back up what the brief says. ## 2. Check what the agent found Start with the release notes. For details such as plan availability or setup steps, ask the agent to check the relevant official help page too. If the sources disagree or a detail is missing, keep that item open for review. If the agent cannot reach a page, give it the official text or another source it can access. A missing source is a reason to flag the claim for review. ## 3. Review the file Open the brief and check its most important claim against the source. Make sure the version is right and the summary says who is affected. If it mentions a new feature, check whether the source says the feature is available to everyone or limited to a particular plan. A reply in chat is useful for seeing what the agent did. Open the actual file to verify that the brief was saved and is ready to share. If the task is still queued, check again once the agent has finished. ## 4. Keep what you will need next time Save the reviewed brief with the release version or date. For the next update, you can ask the same OpenBot agent to read it for the format and check the new release notes for the facts. A previous brief is a useful template. Check its claims again before reusing them in a new version. ## When to move the task into OpenBot Workflows Start by running the task from chat. Once the steps are clear and you need to repeat them, you can build a workflow on OpenBot’s Workflows page. A workflow can include agent steps and an approval step that waits for you. For example, set the agent step to prepare the brief, then connect it to an Approval step. Open the saved brief before you choose Approve or Reject. Connect the step that shares the result after Approval so it runs only after you approve. Run the workflow once and check that you can find the brief and review it before anything is shared. For a single release brief, a clear prompt, a saved file and your review are enough. --- Source: https://openbot.run/guides/ai-teammates-vs-ai-chatbots [All guides](https://openbot.run/guides) # AI Teammates vs AI Chatbots: When You Need More Than One Conversation See when a quick AI chat is enough and how to divide research, writing and review between OpenBot agents that can pass files to each other. Nov 10, 2026 Norbert Bodziony A chat is enough when you need a quick explanation or a second opinion on a paragraph. When the same project needs research, writing and review, it helps to give each part a clear owner. In OpenBot, you can give agents their own roles and pass work between them. AI teammates and AI chatbots can both work through a chat interface. The choice depends on what has to happen after the answer. ## When one chat is enough Say you have an onboarding email and want to make it clearer. Paste the draft, tell the assistant who will read it and ask for a rewrite. You can check the result in a few minutes and move on. If the same agent can handle the whole job and you can easily review its work, keep the setup simple. An ongoing project can also start with one OpenBot agent that keeps its [conversation and workspace](https://openbot.run/guides/ai-workspace-vs-chat) for later tasks. ## When to give the work separate owners Suppose your team maintains a customer onboarding guide. Each product update can change a step, a button name or a plan requirement. You need someone to find those changes, someone to update the guide and someone to check the result. Here is one way to divide that work in OpenBot: 1. **Researcher.** Check the official release notes and help pages. Save a short list of confirmed changes, links and any questions that remain open. 2. **Writer.** Read those findings and update the existing guide. Mark the sections that changed and keep uncertain details out of the draft until they are confirmed. 3. **Reviewer.** Compare the updated guide with the source pages and the original brief. Return a list of corrections for the writer or flag the draft as ready for your review. Try it with two agents first. One can research and draft, while the other checks the result. Add a separate writer if the work regularly calls for that split. ## Give the next agent a usable handoff OpenBot agents can message each other and attach files. Agents on the same OpenBot computer or server can also use its shared folder. A channel gives participating agents a shared conversation for the task. When you ask the writer to pass a draft to a reviewer, include the brief, the draft, the source notes and the checks you want. The reviewer gets the material you pass along. The writer’s private chat and memories are separate from that handoff. For example, a review request could say: > Check the steps in the attached onboarding draft against the links in the research notes. Flag any button names or plan requirements you cannot verify. Return your findings and leave the draft file for the writer to update. Keep one agent responsible for editing the final file. That makes it clear who should apply corrections and which version you should read. ## Choose based on the next step Use a chat for a question or a quick edit. Use one persistent agent for recurring work that needs the same instructions and files. Split the work between agents when there is a clear reason to hand it off, such as checking a draft against its sources before it goes out. If the handoff takes more effort than reviewing the answer yourself, keep the task with one agent. The right setup is the one that gives you a result you can check. --- Source: https://openbot.run/guides/ai-workspace-vs-chat [All guides](https://openbot.run/guides) # AI Agent Workspace vs Chat History: What Changes in Practice? Compare an AI agent workspace with chat history in OpenBot. See how saved files, shared work and model changes affect an ongoing research task. Nov 3, 2026 Norbert Bodziony A long chat can show how a report was written. An AI agent workspace gives you a place to keep the report itself, the brief and the sources behind it. The difference becomes useful when you return to the same job next week. In OpenBot, an agent has both a working folder and a saved conversation. You can discuss the work in chat, ask the agent to save the result, and return to those files for the next task. Here is what that changes in a recurring research project. ## Chat history and workspace files do different jobs Chat history records your requests, the agent's answers and the discussion between them. It helps explain why you chose one option over another. Workspace files give you a clear place for the current brief, source notes and finished report. You can ask the agent to open and update a named file instead of finding the right version among several replies. Chat apps can also offer files and project features. This comparison is about how you organize the work: mainly through messages, or through messages plus clearly named project files. ## An example: updating a software comparison Suppose you ask an agent to compare three tools for your team. You discuss the budget, change the criteria and ask for two revisions. If you leave everything in the conversation, the latest answer may be clear today. Next week, you may need to check which version you approved and which earlier suggestions you rejected. In OpenBot, you can ask the agent to save four files in its workspace: - `brief.txt` — who the comparison is for and what matters. - `sources.txt` — source links, checked dates and facts that need another check. - `report.txt` — the report you are reviewing. - `decisions.txt` — your accepted choices and reasons for excluding an option. These are suggested filenames, not files OpenBot creates automatically. Ask the agent to create them and review their contents before treating them as the agreed project record. ## Returning to the task next week Your next request can point to that record: > Read `brief.txt`, `decisions.txt` and `report.txt`. Recheck the product facts in `sources.txt` against current official sources. Save an updated report as a new dated file, keep the previous report, and list what changed. Flag anything you could not verify. The agent has a clear starting point: the current requirements, the previous result and the decisions you already made. You can review the changes before accepting the new report. OpenBot keeps the conversation too. Saved files give you another way to identify the current result; they do not replace the discussion that led to it. Saving a file does not mean the agent reads every project file on every task. Name the files it should read, especially after a long gap or a change of model. ## Changing the model while keeping the work You can change the AI provider an agent uses, such as moving from Claude to ChatGPT. The agent keeps its name, saved conversation and workspace files. You might use one provider for research and another to revise the same report. The new model does not automatically receive the whole earlier conversation. OpenBot passes along selected parts, and the agent can look up more of its saved history when needed. Some details of the old provider's work do not transfer. A short brief and a saved report help you make the next instruction specific: "Read these files, check the sources and revise this section." ## Passing the report to another OpenBot agent Agents on the same OpenBot computer or server can use its shared folder for files they need to work on together. They can also send messages with files attached. A channel gives the team a place to discuss the shared task. For example, one agent can collect sources, another can draft the report, and a third can review it. The request to the reviewer should include the brief, the report and the checks you want it to make. Sharing a file does not automatically share an agent's entire private conversation or memories. Include the context the next agent needs. Choose one agent to update the main report. Ask the others to return separate research or review files so that several agents do not overwrite the same document. The working files live on the computer or server running that OpenBot instance. Connecting from your phone does not turn its shared folder into a folder synchronized across all your devices. ## What a workspace does not fix A saved source can still be out of date. A report can still contain a mistake. Ask for fresh checks, keep earlier versions you need, and read the result before using or publishing it. A folder also does not decide what an agent can access. OpenBot's Workspace only setting, for supported providers, limits writes to allowed folders; it does not block file reads or network access. Full access is a separate setting. This example describes a workflow you can try, not a measured time saving. Its value depends on whether the files stay clear and current and whether the agent actually uses them. For more tasks to try, see our [AI agent examples](https://openbot.run/guides/ai-agents-examples). ## Frequently asked questions ### Should I stop using chat? No. Use chat for instructions, questions and feedback. Use files for the brief, source notes and results you want to return to. ### What should I save first? Start with the agreed brief and the latest report. Add source notes and decisions when the project needs them. ### Is `report.txt` automatically approved when the agent saves it? No. In this example, approval is your review decision. Ask the agent to keep the previous file and save a new version before making changes you may want to undo. --- Source: https://openbot.run/guides/install-openbot-vps [All guides](https://openbot.run/guides) # How to Install OpenBot on a VPS: Ubuntu and Docker Guide Install OpenBot on a VPS with Ubuntu or Docker. Connect the server to your account, set up an AI provider, and manage updates, logs and backups. Oct 30, 2026 Norbert Bodziony Installing OpenBot on a VPS gives your agents a computer that can stay running when you close your laptop. Their files, conversations and tools live on the server, and you connect from the OpenBot app or your browser. You can install OpenBot directly on Ubuntu or run the official Docker image. Both methods connect the server to your OpenBot account. Choose one, then follow the same steps to open the server and set up your first agent. ## What you need For the Ubuntu method, use a 64-bit VPS running **Ubuntu 24.04 LTS**, with **systemd** and access to **sudo**. OpenBot provides builds for x86_64 and arm64. Ubuntu 24.04 is the baseline used for the full installation tests in the project's documentation. Have your SSH login details ready, plus access to the email address you use for OpenBot. You will also need access to an AI provider when you create an agent. The server installation does not include an AI subscription. Choose enough RAM and disk space for the work your agents will do. Browser sessions, coding tools and several agents can use more memory than a single text task. Monitor the server as you add work; this guide does not set a universal minimum VPS size. ## 1. Connect to your VPS Open a terminal on your computer and connect using the SSH details supplied by your VPS provider: ``` ssh YOUR_USER@YOUR_VPS_IP ``` Replace `YOUR_USER` and `YOUR_VPS_IP` with your actual login and server address. Run the installation commands below inside that SSH session, on the VPS. On a fresh Ubuntu server, install the download tools first: ``` sudo apt-get update sudo apt-get install -y curl ca-certificates ``` ## 2. Choose an installation method | Method | When to use it | | --- | --- | | **Ubuntu installer** | You want OpenBot to run as a system service on the VPS. | | **Docker Compose** | You already manage services with Docker, or prefer to keep OpenBot in a container. | ### Option A: Install directly on Ubuntu Run the official OpenBot server installer: ``` curl -fsSL \ https://raw.githubusercontent.com/nightly-labs/openbot/main/scripts/install-server.sh | sudo bash ``` The installer downloads the latest released Linux build, verifies its checksum and sets up the OpenBot service. On a fresh installation, it creates a separate `openbot` service user and starts OpenBot in the background. Sign in and give the server a name you will recognize in the app: ``` sudo openbot login --name "OpenBot VPS" ``` Enter the email address for your OpenBot account, then the code sent to it. Use the same account when you open the app later. Check the result: ``` sudo openbot status ``` Look for your installed version, account and server state. If the command cannot reach OpenBot yet, check the service log using the troubleshooting section below. OpenBot runs as a system service, so closing the SSH session does not stop it. Continue to step 3. ### Option B: Run OpenBot with Docker Compose This option needs Docker Engine and the Docker Compose plugin on the VPS. If they are not installed, follow the [Docker installation instructions for Ubuntu](https://docs.docker.com/engine/install/ubuntu/) first. Create a new folder for this deployment. Keep its Compose file and security profile together: ``` mkdir openbot-vps cd openbot-vps curl -fsSLO \ https://raw.githubusercontent.com/nightly-labs/openbot/main/docker/compose.yaml curl -fsSLO \ https://raw.githubusercontent.com/nightly-labs/openbot/main/docker/seccomp.json sudo docker compose up -d ``` Use another new folder name if `openbot-vps` already exists. The official Compose file uses the latest release image and a persistent data volume. It also includes the security profile and shared-memory setting required by the container. Keep those settings in place. The Electron sandbox stays enabled; the supplied security profile lets it run in the container. Agent commands and files live inside that container, so mount only the host folders you intend agents to use. Sign in to OpenBot and check the server: ``` sudo docker compose exec openbot openbot login --name "OpenBot VPS" sudo docker compose exec openbot openbot status ``` Enter your OpenBot account email and its sign-in code. Run later Compose commands from this same deployment folder. The container uses a persistent volume for its data and restarts according to the Compose policy. Continue to step 3. ## 3. Open the server in OpenBot Open the OpenBot desktop app or [OpenBot in your browser](https://openbot.run/app) and sign in to the same account. Find **OpenBot VPS** in the server list and select it. The standard setup connects through OpenBot's existing account and remote-access services. You use the app to reach the VPS; the OpenBot container publishes no web port of its own. When you work on this server, your agents' workspace files and conversations are stored on the VPS. AI providers, websites and connected services can still receive the task data needed for their work. ## 4. Connect a provider and create an agent Choose the provider you want to use on the server and follow its sign-in or API-key flow in OpenBot. A provider connection on your laptop is not automatically a working connection on the VPS. If this is a new server with no agents, OpenBot offers provider setup before the first-agent form. Once connected, create an agent, give it clear instructions and start with a small task. For example: > Create a file called server-check.txt in your workspace. Put today's date and a short description of what you can access in it. Do not change system settings or send anything externally. Open the file in the agent's workspace and check the result. The [first-agent guide](https://openbot.run/guides/openbot-101) walks through the agent setup in more detail. Your VPS can stay available around the clock while its service, network and provider access are working. Check routines and their times before leaving scheduled work running. ### Use Hermes Agent on the server Open the server's provider settings and select Download on the Hermes row. OpenBot installs Hermes Agent for its service user, then the row shows Connect. To install Hermes from a terminal instead, install it as the service user, not as your SSH login. For the Ubuntu installer, run this on the VPS: ``` sudo -u openbot -i bash -c \ 'curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --non-interactive' sudo openbot restart ``` For Docker, install it in the container. Hermes stays in the data volume when you replace the container: ``` sudo docker compose exec openbot bash -c \ 'curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --non-interactive' sudo docker compose restart openbot ``` Then select Connect on the Hermes row and enter the Nous Portal code. You can also run `hermes setup` or `hermes model` as that user to choose another model provider. ## 5. Read logs and keep the server updated For the direct Ubuntu installation, these are the main commands: | Task | Command | | --- | --- | | Check the server | `sudo openbot status` | | Follow its logs | `sudo openbot logs -f` | | Restart OpenBot | `sudo openbot restart` | | Install the latest update | `sudo openbot update` | An update restarts OpenBot, so run it when an interruption is acceptable. Back up the server data first. The native service also checks for updates and can stage them for installation; `openbot status` shows a staged update. For Docker Compose, use these commands from your deployment folder: ``` sudo docker compose logs --tail 100 openbot sudo docker compose exec openbot openbot status ``` To upgrade after checking the release notes and making a backup: ``` sudo docker compose pull sudo docker compose up -d sudo docker compose exec openbot openbot status ``` A container upgrade replaces the image and keeps its data volume. The `latest` tag follows new releases. If you want to choose a specific release, set its version tag in `compose.yaml` before upgrading. Use the same volume and runtime settings. OpenBot can migrate its database during an upgrade. Use a compatible release with your backup when recovering; switching an upgraded live volume to an older image is not a safe rollback. ## Keep a backup of the complete setup For the native installation with the default service user, keep the service user's home at `/var/lib/openbot` and the keyring data at `/srv/openbot-hosted` together. If you selected another service user, back up that user's actual home instead. Stop OpenBot before making a consistent copy, then start it again when the copy is complete: ``` sudo openbot stop ``` Back up both locations using your normal backup process, then run: ``` sudo openbot start ``` For Docker, stop the service and back up the complete persistent data volume, then start it again. The volume includes provider sign-ins, the database and its encryption keyring. A copy of the workspace files alone is not a complete backup. Protect backups like account credentials and keep a copy outside the VPS. A persistent disk or Docker volume can survive a restart, but it is still part of the server you manage. ## If the server does not appear Check that you completed `openbot login` on the VPS and that the app uses the same email account. A successful installation alone does not sign the server in. For the native install, read `sudo openbot status` and `sudo openbot logs -f`. For Docker, check `sudo docker compose ps` and the container logs. Follow any reported errors before trying to reinstall. If the container reports a user-namespace or sandbox problem, check that it uses the official security profile. Some VPS hosts restrict user namespaces or apply extra AppArmor rules. Keep the sandbox enabled and follow the [container documentation](https://github.com/nightly-labs/openbot/blob/main/docs/docker.md) for those host limits. If work starts waiting when memory is low, inspect the memory information in `openbot status`. Reduce concurrent work or give the VPS more memory for the tasks you want to run. ## Frequently asked questions **Will agents keep working when I close my laptop?** Yes, while OpenBot is running on the VPS and the required services remain available. Your laptop is the client; the VPS runs the agents. **Do I need a domain for this setup?** The standard setup uses OpenBot's existing account services and app. A custom domain is not required for this installation. **Does the VPS include model access?** You connect a provider in OpenBot. VPS costs and provider access are separate. **Can I move my existing agents to the VPS?** Yes. If you administer both servers, open **Transfer** in **Server settings** in the desktop app and choose **Copy to another server**. Pick the server that has the agents and choose this VPS as the destination. You can copy the agents in one step without handling a file. ZIP import still works. To read a local Hermes folder, use the OpenBot desktop app on the computer where the folder is stored. --- Source: https://openbot.run/guides/import-hermes-agents [All guides](https://openbot.run/guides) # How to Import Hermes Agents into OpenBot Import your Hermes Agent profiles into OpenBot. Choose what to bring over, review instructions, memories and skills, and check your agents after import. Oct 27, 2026 Norbert Bodziony You can move your Hermes Agent setup into OpenBot without rebuilding every agent from scratch. Each Hermes profile becomes an OpenBot agent, with its own instructions, saved memories and custom skills where available. OpenBot shows you what it found before you import. You choose the profiles to bring over, check which items will move and allow any optional items you want to include. Your original Hermes profiles stay where they are. ## Before you start Update OpenBot and decide where your imported agents should live. For the folder method below, use the OpenBot desktop app on the same computer as Hermes and select that computer as your destination. If Hermes is on another computer, or you use OpenBot in the browser or on a joined server, use the ZIP export method later in this guide. ## 1. Open Import in OpenBot Open **Server settings → Import**, then choose **Hermes Agent**. ![OpenBot Import settings with Hermes Agent, Grok Bot and another OpenBot computer as source options.](https://openbot.run/assets/import-source-BMWDElDp.png) Choose Hermes Agent in the Import settings. Choose the option that matches where Hermes runs and how you want to prepare the import. ## 2. Choose an import option | Option in OpenBot | When to use it | | --- | --- | | **Read this computer** | Hermes is on the same computer as the OpenBot desktop app. No ZIP is needed. | | **Install with a command** | Add the export skill in Hermes with one command, then import its ZIP into OpenBot. | | **Set it up yourself** | Save the supplied skill file in Hermes yourself, then use the same ZIP export flow. | For the two ZIP options, prepare the export on the computer where Hermes runs. You can then import it through the OpenBot desktop app, the browser or a joined server. ### Option 1: Read this computer Select **Read the Hermes folder**. OpenBot reads the profiles it finds without changing the Hermes folder. ![Hermes import controls with Read this computer selected and the Read the Hermes folder button.](https://openbot.run/assets/read-folder-DM2qYxbm.png) Read this computer: open the local Hermes profiles directly. The usual folder is `~/.hermes`. If OpenBot cannot find your profiles, use **Choose the Hermes folder** and select the folder your installation uses. On the profile preview, **Choose another folder** lets you change it. You do not need to make a ZIP for this method. OpenBot takes you straight to the profiles it found. Once the folder has been read, continue to step 3 to review the profiles. ### Option 2: Install with a command This option adds the OpenBot export skill to Hermes. It is useful when Hermes runs on another computer, or when you want to import through the browser or into a joined OpenBot server. In **Server settings → Import → Hermes Agent**, choose **Install with a command**. Copy the command into a Hermes chat: ![Hermes import controls with Install with a command selected, showing the skill installation command.](https://openbot.run/assets/install-command-BX5laHbT.png) Install with a command: add the export skill from a Hermes chat. ``` /skills install nightly-labs/openbot/resources/agent-import/hermes-agent ``` Hermes scans the skill before installing it. If you prefer the terminal, use: ``` hermes skills install nightly-labs/openbot/resources/agent-import/hermes-agent ``` After installing the skill, follow the ZIP export steps below. ### Option 3: Set it up yourself Use this option if you prefer to add the export skill as a file. On the Hermes Agent import page in OpenBot, choose **Set it up yourself**. ![Hermes import controls with Set it up yourself selected and Copy and Save file actions for SKILL.md.](https://openbot.run/assets/manual-setup-BzZT9AFT.png) Set it up yourself: copy or save the export skill file in Hermes. Select **Save file…** to save the supplied `SKILL.md`, or use **Copy** and paste its complete contents into a file with that name. Save it at `~/.hermes/skills/openbot-export/SKILL.md`. If your Hermes installation uses a different home folder, put it in that folder under `skills/openbot-export/SKILL.md`. Create the `openbot-export` folder if needed. Start a new Hermes chat so it can find the skill. The export leaves this helper skill out of the ZIP. You do not need to run the installation command as well. ### Export a ZIP for options 2 and 3 Both **Install with a command** and **Set it up yourself** continue with the same export step. In Hermes, send: ``` /openbot-export ``` Choose the profiles you want to export and wait for the completed ZIP in Downloads. Its name will look like `openbot-export-hermes-2026-10-28.zip`. This export can carry your persona, saved memories, custom skills and supported recurring jobs. It leaves out credentials, chat sessions, bundled skills and MCP server settings. The ZIP export skill also leaves out one-time jobs; read its summary of skipped items. Return to OpenBot, select **Choose export file**, and choose the completed ZIP without unzipping it. OpenBot opens a preview. Continue to step 3 to review it before importing. ## 3. Review what will move You can review the import before OpenBot adds the agents. The preview depends on whether you read the Hermes folder or selected a ZIP. ### If you read the Hermes folder Use the checkbox beside each profile to choose which agents you want in OpenBot. You can start with one profile and bring over the others later. The default profile appears as **Hermes**; named profiles appear separately. Expand the item count beside a profile, such as **4 items**, to see its contents. The labels explain what will happen: - **Imports** means the item will come with the selected profile. - **Allow** means the item needs a separate choice before it can be included. For example, a custom skill that runs code needs your approval. - **Not imported** means the item will stay behind. Read the reason shown under it. In the example below, `SOUL.md` will become the agent's instructions. The built-in Hermes skills, secrets and past sessions are marked **Not imported**. Your list may contain different items and counts. ![One selected Hermes profile with SOUL.md marked Imports, built-in skills, secrets and past sessions marked Not imported, and an Import 1 agent button.](https://openbot.run/assets/review-profiles-Do7-HxeC.png) Select the profiles you want, then check what imports and what stays in Hermes. If a profile has saved memories, OpenBot can bring over entries from `MEMORY.md` and `USER.md`. Custom skills can come too; skills bundled with Hermes are left out. If the profile only has Hermes's starting persona text, the new agent keeps OpenBot's starting instructions. Supported scheduled jobs become OpenBot routines. Check their schedules and enabled state before importing: enabled routines can remain enabled. Jobs that only run a script and unsupported schedules are left out with a reason. If the list includes MCP servers, these are connections to tools and services. You can allow supported connection settings to be added. They arrive turned off, with sensitive connection values cleared, so you can check them before use. ### If you selected a ZIP Select the agents you want in the ZIP preview. Check the counts of included memories, skills, routines and files, and read any warnings about items that will not move. A ZIP import creates separate agents. If a name already exists, OpenBot asks you to confirm the duplicate when you start the import. It does not offer the folder method's linked-profile update. ## 4. Import the selected agents Check your selected agents and any items you allowed, then select the Import button. Its label shows your selection, such as **Import 1 agent**. When the process finishes, OpenBot shows **Import finished** and lists the imported agents. Read any warnings or **Not imported** results, then open an agent from the list or select **Done**. ![Hermes Agent import result showing Import finished, 1 agent imported, the Hermes agent and a Done button.](https://openbot.run/assets/import-finished-BkDt9GDl.png) The result screen lists the agents imported into OpenBot. If you read the same Hermes folder again later, OpenBot recognizes profiles it imported before. **Update it** updates the linked agent while keeping its OpenBot chats. **Leave it** keeps that agent as it is. ## 5. Check the agent before its first task Open the imported agent and check its instructions, memories, skills and routines. Choose the provider and model you want it to use in OpenBot, then complete any sign-in it needs. If a custom persona was too long for the instructions field, folder import saves the full version in `imported/hermes/SOUL.md` and leaves a shorter instruction pointing to it. If you allowed MCP settings, check them in OpenBot in Marketplace, Apps, supply any required values and enable the connection when ready. Messaging connections and Hermes terminal backends also need to be set up for the way you work in OpenBot. Project documents and a complete working directory are not part of the standard Hermes profile import. Bring any files your agent needs separately. A useful first request is: > Check your instructions, saved memories and installed skills. Tell me which tools and files are available and which connections still need setting up. Do not send messages, publish content or change project files. ## What stays in Hermes? Full chat history stays in Hermes. OpenBot brings over saved memory entries, subject to its memory limits. Read any warnings to see what could not be included. Credential files and login sessions are not transferred. Reconnect providers and services in OpenBot. Also check any sensitive information you may have written into your own instructions or memories. For the folder method, the reader leaves the source profiles unchanged. For a ZIP import into a joined server, the file goes through the host connection to the computer running that server. Check the selected destination before importing. ## Frequently asked questions **Will each Hermes profile become a separate agent?** Yes. Select the profiles you want on the preview screen. **Can I keep the same personality and memories?** Your custom persona and saved memory entries can be brought over. A different model or set of tools can still change how the agent behaves, so start with a small task. **Will all tools work immediately?** Some items need your approval or a new connection. MCP settings imported from a folder arrive turned off. Provider sign-ins, credentials and messaging connections need to be set up again. --- Source: https://openbot.run/guides/import-grok-bot-agents [All guides](https://openbot.run/guides) # How to Import Grok Bot Agents into OpenBot Move your Grok Bot agents into OpenBot with one ZIP export. Follow the steps for instructions, skills, memories, routines and optional files. Oct 23, 2026 Norbert Bodziony You can bring your Grok Bot agents into OpenBot without rebuilding their setup one by one. An export agent in Grok Bot packs the agents you choose into a ZIP file. OpenBot reads that file, shows a preview and creates the selected agents. The import can carry instructions, avatars, skills, routines, memories and optional workspace files. Group chats can become OpenBot channels. Original chat transcripts are not transferred: the export agent saves important facts from them as memories instead. ## Before you start Update OpenBot and decide where the agents should live: on this computer or on an OpenBot server you have joined. The import goes into the server whose settings you open. You will also need access to your agents in Grok Bot and somewhere to save the export file. You can keep using the originals while you check the copies in OpenBot. ## 1. Open Import in OpenBot Open **Server settings → Import**, then choose **Grok Bot**. ![OpenBot Import settings showing Grok Bot as an import source.](https://openbot.run/assets/import-source-BMWDElDp.png) Choose Grok Bot as the import source in OpenBot. The page shows the export instructions and a card for **OpenBot export**. Select **Open the export agent**. This opens the Grok Bot page for **Migrator by OpenBot**, the agent prepared for this job. Add it to Grok Bot. It is a separate export agent; you do not need to replace the instructions of an agent you already use. ![Migrator by OpenBot in Grok Bot, with the Add Bot button.](https://openbot.run/assets/add-migrator-Bd1TZ7fv.png) Add Migrator by OpenBot to Grok Bot. ## 2. Ask the export agent to prepare your agents Send this message in the export agent's Grok Bot conversation: > Export my agents for OpenBot ![OpenBot instructions for adding the Grok Bot export agent, requesting an export and choosing the ZIP file.](https://openbot.run/assets/export-instructions-B5cr8NXR.png) Use the export message shown in OpenBot, then return with the completed ZIP file. It should list the agents and group chats it finds, then ask which ones you want and whether to include workspace files. Choose a small set for your first import if you would like to check the result before moving the rest. Including files brings the selected working documents along. Leaving them out makes a smaller export. The export skill tells the agent to exclude credential files, passwords, tokens and other secrets; keep these out of any material you include. The export agent leaves itself out of the archive. It also leaves out its own export skill. ## 3. Save the ZIP file When the export finishes, the agent should give you one file in **Downloads**, named like `openbot-export-2026-10-27.zip`. The date in the name depends on when you export it. Use the completed ZIP file. Do not unzip it or select a temporary `.zip.part` file. Read the export agent's summary of what it included and anything it could not read. If an important agent or file is missing, ask it to correct the export before continuing. ## 4. Review the export in OpenBot Return to **Server settings → Import → Grok Bot** and select **Choose export file**. OpenBot shows the agents before importing them, with counts for items such as skills, routines, memories and files. If the export includes group chats, review the channels too. A channel needs at least one of its agents selected to be imported. Select the agents and channels you want. Read any warnings about items that will not move, then use the Import button showing your selected counts. If a matching agent name already exists, OpenBot asks you to confirm the duplicate. A ZIP import creates a separate agent; it does not update the existing one. ## 5. Check the result The **Import finished** screen lists what was imported and anything that was skipped. Open an imported agent and check its instructions, memories, skills and routines. Workspace files included in the export are placed in an `imported/` folder inside that agent's OpenBot workspace. If the original instructions were long, the export may keep a short version in the instructions and put the full text in `imported/INSTRUCTIONS.md`. Choose the provider and model you want to use in OpenBot, and complete any sign-in it requires. Moving an agent's setup does not transfer a Grok Bot subscription or sign-in session. Review enabled routines and their times before relying on them. Then give the agent one small task, such as: > Read your imported instructions and memories, list the files you can find in imported/, and tell me what needs setting up before you continue my work. Do not send or publish anything. ## What moves, and what stays behind? The export is designed to carry the agent's name, role, instructions, avatar, skills, supported routines and memories. Selected working files are optional. Group-chat setup, membership and saved facts can become channels and channel memories. The original conversations do not become OpenBot chat history. Memories are a summary of selected facts, so check that the preferences, decisions and unfinished work you care about are represented correctly. Provider access and external services may need to be connected again. For an import into this computer, OpenBot reads the ZIP locally. For a joined server, the ZIP is sent through the encrypted host connection to the computer running that server. If a remote export is too large, try fewer agents or leave workspace files out. ## If the import does not work If you cannot find the Grok Bot source, check that OpenBot is up to date and that you opened the settings of the intended server. If the ZIP is rejected, ask the export agent to rebuild it using the OpenBot export skill. It needs the OpenBot import format; a general backup of the Grok Bot workspace is not the same file. If only part of an import succeeds, read **Not imported** and the warnings before trying again. Select the missing agents on a retry so you do not create extra copies of agents that already arrived. ## Frequently asked questions **Will I keep all my chat history?** No. The export agent turns important facts into memories. Original messages and their full conversation history are not imported. **Can I move only one agent?** Yes. Choose it when preparing the export or select only that agent in OpenBot's preview. **What if I cannot install the ready export agent?** OpenBot also offers **Set it up yourself**. Copy or save the supplied `SKILL.md`, add it to a separate export agent in Grok Bot, then send the same export message. --- Source: https://openbot.run/guides/ai-agents-examples [All guides](https://openbot.run/guides) # 20 Real AI Agent Examples for Work and Business Explore 20 practical AI agent examples for research, content, coding and daily operations. Each OpenBot task ends with a result you can review. Oct 20, 2026 Norbert Bodziony The best AI agent examples begin with a task you already do and end with a result you can check. That might be a research brief, a draft email, a proposed code change or a cleaned spreadsheet. You do not need to hand an agent the whole business to make it useful. In OpenBot, you can give an agent instructions, let it work with files and approved web pages, and return to the same conversation later. You can also pass work to another agent when a separate reviewer or writer helps. The 20 examples below are tasks to try, not customer case studies or promises that every result will be correct. ## Research and knowledge 1. **Compare vendors.** Ask a research agent to read three official pricing pages and save a table with prices, limits and links. Check the dates and any unclear terms before choosing a plan. 2. **Track documentation changes.** Give an agent last month's saved documentation and the current pages. Ask for a short list of additions, removals and changed instructions, with links you can open. 3. **Prepare for a meeting.** Supply the agenda and your notes. Ask the agent for a one-page brief with open questions. Check that it labels guesses as guesses. 4. **Answer a question from approved files.** Give the agent a policy or product guide and ask it to point to the passage behind each answer. If the files disagree, the result should show the conflict. 5. **Build a source brief.** Ask the agent to collect claims, dates and source links about one topic. A writer can then use the saved brief without relying on an unsupported summary. ## Content and marketing 6. **Make a content brief.** Provide the audience, topic and approved keyword list. Ask for an outline, reader questions and a source plan. Do not let the agent invent search-volume numbers. 7. **Refresh an old article.** Give the agent the article and current release notes. Ask it to propose only the paragraphs that need an update and to list the facts you should check. 8. **Turn one approved article into other drafts.** Ask for a newsletter introduction and two social posts. Read them before publishing; short versions can lose an important caveat. 9. **Check brand language.** Save your style guide in the agent's workspace. Ask it to flag terms that do not match the guide and suggest exact edits, not a vague score. 10. **Summarize a campaign export.** Give the agent an approved CSV file and define the metrics you need. Ask for a short report plus the calculations, then compare its totals with the file. ## Software and product work 11. **Investigate a bug.** Give a coding agent the issue and access to the relevant repository. Ask it to reproduce the problem and save its findings before changing code. 12. **Review a proposed code change.** Ask another agent to inspect a pull request and report specific problems with file locations and a failure scenario. A person still decides which comments are valid. 13. **Update instructions after a code change.** Ask the agent to compare a merged change with setup documentation and propose a documentation patch. Run any changed steps yourself before publishing them. 14. **Draft release notes.** Give the agent the merged changes and your writing style. Ask for notes that separate what has shipped from what is still planned. 15. **Plan a dependency update.** Ask the agent to read the official migration guide, find affected code and list likely tests. Review the plan before it installs or changes anything. ## Daily operations 16. **Draft a support answer.** Supply the customer's question and approved help material. Ask for a reply draft with any missing account details clearly marked. A support owner checks it before sending. 17. **Find possible duplicate contacts.** Give the agent a CRM export, not write access to the live CRM. Ask for a list of record IDs that may match and the reason for each suggestion. You decide which records to merge. 18. **Write a weekly project update.** Provide issue and document exports. Ask the agent to separate completed work, open work and items it could not verify. 19. **Check a data file.** Ask the agent to find missing values, duplicate rows and dates in the wrong format. Keep the original file and review the proposed cleaned copy. 20. **Report unusual changes.** Give the agent this week's figures and an accepted baseline. Ask it to show the biggest differences and say “no significant change” when that is the honest result. ## Which example should you try first? Pick a task whose answer you can check in a few minutes. Vendor research, an article update or a report from a small export are good starts. Give the OpenBot agent the input, the result you want, the sources it may use and the actions it must leave to you. Ask it to save a file and mark anything it cannot confirm. Start with one agent. If you later need a second opinion, another OpenBot agent can review the saved file or take a separate writing task. Your review remains the final step for important facts, messages, code changes and publishing. If you have not created an agent yet, start with [How to Use OpenBot](https://openbot.run/guides/openbot-101). ## Frequently asked questions ### Do I need 20 agents for these tasks? No. One agent can handle several related tasks. Add a specialist when its role makes the work easier to check. ### Can an agent do all of this without my review? It may do many steps, but important results still need a person to check facts and approve actions such as sending, merging or publishing. ### How do I know whether a task helped? Look at the accepted result, how much you had to correct and whether the agent showed its sources and uncertainties. A long activity log is not the same as useful work. --- Source: https://openbot.run/guides/ai-agents-vs-agentic-ai [All guides](https://openbot.run/guides) # AI Agents vs Agentic AI: What Is the Difference? AI agents vs agentic AI, explained with a simple OpenBot example. See what an agent is, what makes its work agentic, and when one agent is enough. Oct 16, 2026 Norbert Bodziony An AI agent is something you can give a job to. [Agentic AI](https://openbot.run/guides/agentic-ai-definition) describes how it works when it can take steps toward that job, use tools and adjust its next step based on what it finds. The terms sound similar because they describe two sides of the same task: the worker and the way the work gets done. OpenBot makes the difference easy to see. You can create a named agent, give it instructions and keep its conversations and files in its own workspace. What happens after you assign a task shows how much of its work is agentic. ## One OpenBot task, two terms Suppose you want to compare three project-management tools for your team. You ask an OpenBot research agent to check their official pages and save a short comparison with links. The research agent is the AI agent. It has a role, instructions and a place to keep its work. Now imagine that one pricing page does not explain a team limit. The agent opens the vendor's help page, checks the missing detail and marks anything it still cannot confirm. Choosing that next step from what it found is agentic behavior. You should still read the comparison before making a purchase. Agentic AI can move the research forward; it does not make every finding correct. ## Does every AI agent work this way? Not to the same degree. One agent might follow a set sequence: read these pages, fill in this table and stop. Another might decide which source to check next when the first page is incomplete. Both may be called AI agents, but the second has more room to adapt. The label alone tells you very little. Ask what the agent can actually do: Can it open the sources you approve? Can it save a file you can inspect? Will it show where a claim came from and say when it cannot verify one? Can you review its result before anything is sent or published? In OpenBot, the agent's workspace and conversation stay on the computer running OpenBot. A connected AI provider may still receive the prompts needed to carry out the task, and the browser and other tools may use the network. Those details matter more than whether a product calls itself “agentic.” ## Do you need several agents? No. One OpenBot agent can research and save the comparison. If the work grows, you can ask a second agent to review the evidence or turn the approved notes into a report. OpenBot supports messages, shared files and handoffs between agents, but adding more agents does not automatically improve the answer. Use separate agents when their jobs are genuinely different. Keep a clear owner for the final result and let a person review important decisions. ## A small task to try Give one OpenBot agent a narrow request: “Compare these three approved product pages. Save a table with the price, team limit and a link for each fact. If a detail is missing, mark it as unknown. Do not send or publish the result.” Then read the file. If the agent found a missing detail and chose a useful next source, you have seen agentic behavior in practice. If it guessed, update the instructions and review process before trusting it with a larger task. ## Frequently asked questions ### Is agentic AI a special model? Usually the term describes a way of working with a model, instructions and tools. It does not tell you which model the product uses. ### Can a single agent be agentic? Yes. It can choose its next step based on what happens. A team of agents is not required. ### What is the useful distinction? “AI agent” names the worker. “Agentic AI” describes the ability to keep working toward a goal across steps. Judge both by the output you can check, not by the label. --- Source: https://openbot.run/guides/agentic-ai-definition [All guides](https://openbot.run/guides) # Agentic AI Definition: How It Differs From Chatbots What is agentic AI? Compare OpenBot agents with a typical chatbot: ongoing tasks, saved files, source checking and handoffs between agents. Oct 13, 2026 Norbert Bodziony Agentic AI means AI that can work toward a goal through several steps. It can use a tool, look at the result, and decide what to do next. A chatbot often feels different: you ask a question, get an answer, and decide what to ask next yourself. The easiest way to see the difference is to compare a one-off chat with an AI agent in OpenBot. OpenBot lets you create a named agent, give it instructions, and assign work that may involve a browser, files, or another agent. ## The Same Request, Two Ways of Working Suppose you need a short article about your latest product update. In a typical chatbot conversation, you paste a few links and ask for a summary. Then you ask it to turn the summary into a draft. If a claim looks doubtful, you ask another question, find the source, and move the text into your document. Next week you may need to explain the job again. In OpenBot, you could give a research agent one clear assignment: “Read these release notes, check each important claim against its source, and save a short brief with links. Flag anything you cannot confirm.” The agent can open pages in OpenBot's built-in browser and work with files. You can then pass the brief to a writing agent, or ask the same agent to draft the article. You still read the result and decide whether it is ready to publish. The difference is that the agent can carry out more of the steps between your request and that review. ## What Makes the OpenBot Agent Useful? **It keeps a place for ongoing work.** Each OpenBot agent has its own instructions, conversation, and workspace. If you return to a task later, the conversation and files remain on the computer running OpenBot. You do not have to recreate the same agent every time. **It can leave you a real deliverable.** An answer in chat can be helpful, but a saved brief or draft is easier to review, edit, and pass along. An OpenBot agent can use files and a built-in browser as part of the job. **It can work with another agent.** A research agent can share its notes with a writing agent. A reviewer can check the draft before it reaches you. OpenBot supports messages, file sharing, and task handoffs between agents. Start with one agent; add another when the roles are genuinely different. **You can choose the AI service.** OpenBot can run agents with providers including Codex, Claude Code, Gemini, and Grok. OpenBot organizes the work; the connected AI service helps the agent carry it out. These are practical advantages for a repeated, multi-step job. They do not make every agent answer correct. A missing source should be marked as missing, not turned into a confident sentence. ## Is a Chatbot Always Less Capable? No. “Chatbot” describes how you talk to a product. Some chatbots also browse, use tools, remember information, or run agents. The useful comparison is between a simple, one-off conversation and an agent you can give an ongoing job. For a quick definition, a rough idea, or a short rewrite, chat may be all you need. For research, drafting, checking sources, and handing work between people or agents, OpenBot gives the job a place to continue. For a broader introduction, read [What Are AI Agents?](https://openbot.run/guides/what-are-ai-agents) ## Try a Small Task First Create one OpenBot agent for a narrow job. For example: “Check these three approved pages and save a one-page brief with links. Tell me when a fact cannot be verified. Do not publish anything.” Read the brief, check its sources, and adjust the instructions before giving the agent a larger task. OpenBot agents can read and change files and use the network, so give them tasks you trust and review important results. The AI provider you connect may receive the prompts sent to it, even though OpenBot keeps the workspace and files on the computer running the app. ## Frequently Asked Questions ### What Is an AI Agent in OpenBot? A saved AI teammate with instructions and its own place for conversations and files. You can give it a task and review what it produces. ### Does Agentic AI Mean Publishing Without Approval? No. The agent can do work between steps, while you decide whether a draft is ready to publish. ### Do I Need Several Agents? No. One agent can research and draft. Add a second when a separate review or writing role makes the result easier to manage. --- Source: https://openbot.run/guides/openbot-hosted-servers [All guides](https://openbot.run/guides) # OpenBot Hosted Servers: Plans, Pricing, and How They Work Compare OpenBot hosted server plans, pricing, storage, and team limits. Learn how to set one up and when self-hosting may suit you better. Oct 7, 2026 Norbert Bodziony Running OpenBot for a team on your own computer or VPS gives you control, but you’re also responsible for maintaining that machine. An OpenBot hosted server gives your team a Linux machine to work on, without requiring you to set up a VPS yourself. This guide covers the available plans, how to get started, and when self-hosting might make more sense. ## What Is an OpenBot Hosted Server? An OpenBot hosted server is a Linux machine where your team’s OpenBot runs. OpenBot sets it up for you, so you don’t need to prepare your own computer or VPS. Your agents’ workspaces, conversations, and files stay on the server, and you can connect to your team from another device. A new hosted server can start on a virtual server that OpenBot rents from Leaseweb. The privacy notice names two locations: Amsterdam, and near Washington, D.C. for a purchase in US dollars or Brazilian reais, with Amsterdam as the fallback. When no Leaseweb server is free, the server starts in a boat sandbox, as before. The server provides a place for OpenBot to run; it doesn’t include AI model access. You connect a provider separately, and your agents use the models available through that provider. Prefer a Mac as your server? That’s an option too. We can run OpenBot on a dedicated Mac mini. Get in touch, and we can work out the details together. ## OpenBot Hosted Server Plans The plans differ mainly in how many team members can join, how much storage is available, and how quickly the server runs—not in the number of agents you can create. Starter is for one person, with 25 GB of storage and base speed. It could suit a founder or a freelancer who wants agents to keep working on research, routine admin, or client work while the laptop is closed. Standard supports up to 5 members, with 75 GB of storage and twice the speed of Starter. Think of a small software consultancy working across several client projects: its team might use agents for code review, research, and preparing project notes. Pro supports up to 10 members, with 150 GB of storage and four times the base speed. It’s aimed at larger teams or heavier workloads, especially those involving more browser-based work. You can pay monthly or yearly. The yearly option saves 20%. ![Starter, Standard, and Pro plans with monthly prices in US dollars, billed yearly.](https://openbot.run/assets/pricing-B838MCla.png) Prices are in US dollars and billed yearly. Starter has 25 GB of storage, Standard has 75 GB, and Pro has 150 GB. Work at a company also needs a commercial license. A Starter license for one user costs €10 a month, and a Team license costs €25 a month for two users and €15 for each extra user, up to 10 users. With a hosted server, the license costs 10% less, and you can buy both with one payment. Yearly billing also takes 20% off the license. ## Getting Started with a Hosted Server Once you choose a plan and complete payment, OpenBot provisions a Linux server and connects it to your app. It then takes you straight to provider setup, so you don’t need to look for it in Server settings. Connect an AI provider before creating your first agent. The server plan covers the machine OpenBot runs on; model access comes from the provider you connect. After that, you can set up agents for your team and invite teammates to join. ## What Happens When You’re Not Using the Server? A Leaseweb server runs all the time: it does not stop after the last use. When no Leaseweb machine is free, the server starts in a boat sandbox, as before. That sandbox does not need to stay active when nobody is using it. After a period of inactivity, it stops but keeps its data. It starts again when someone connects, and it can wake shortly before a scheduled routine so the routine runs at the right time. Messages and webhooks wake the sandbox too. When someone mentions OpenBot or messages it in Slack, Discord or Telegram, the sandbox starts and answers once it is back online. A webhook routine works the same way: the first delivery gets a “try again later” answer and starts the sandbox, which takes a few minutes. A later try reaches it. Most webhook senders, such as Shopify and Stripe, try again on their own for hours. ## Hosted Server or Self-Hosting? Many teams wonder whether to use an OpenBot hosted server or run one themselves. The main difference is who provides and looks after the machine. Here’s what to consider when choosing between the two. With a hosted server, OpenBot provides the Linux machine, so your team doesn’t need to set up or maintain its own server computer. If you self-host, OpenBot runs on a machine you control. You have more control over the hardware, but keeping it available and maintaining it are up to you. A hosted server suits teams that want to use OpenBot without looking after a machine. Self-hosting makes more sense if you already have suitable hardware and want to manage the setup yourself. See our guide to [running the team server yourself](https://openbot.run/news/run-the-team-server-yourself) for the self-hosting details. ## In Short If you want a team server without setting up and maintaining the machine yourself, an OpenBot hosted server handles that part. If you’d rather use hardware you control and manage the setup, self-hosting is another option. --- Source: https://openbot.run/guides/openbot-marketplace [All guides](https://openbot.run/guides) # How to Use the OpenBot Marketplace: Agents, Skills, and Plugins Browse the OpenBot Marketplace, install agents, plugins, and skills, and submit your own agent for review. Sep 29, 2026 Norbert Bodziony A useful agent takes more than a name. It needs instructions for the work it does, and it may have skills or routines too. The OpenBot Marketplace lets you add an agent someone else has shared, find plugins and skills for your own agents, or submit an agent you have built. ## Browse the Marketplace In the desktop app, click the puzzle-piece icon at the top of OpenBot’s sidebar. The Marketplace has three tabs: **Agents**, **Plugins**, and **Skills**. You can browse the listings by category or search for a creator or item by name. Open a listing to see what it does before adding it. ![OpenBot Marketplace with the Agents tab selected and the Plugins and Skills tabs visible. Agent listings are grouped by category.](https://openbot.run/assets/marketplace-QiGM5DP3.png) Agents, plugins, and skills each have their own Marketplace tab. ## Add an Agent from the Marketplace In the **Agents** tab, open a listing and read its instructions. They tell you what the agent is meant to do and how it approaches the work. A listing may also include skills (playbooks for particular tasks) and routines (tasks the agent can run on a schedule). ![The Code Review Partner agent listing, showing its instructions, skills, and Install agent button.](https://openbot.run/assets/agent-detail-VRo-PPXs.png) Read the agent’s instructions and check its included skills and routines before installing. If the setup fits, select **Install agent**. OpenBot adds a copy of the agent setup, not the creator’s conversations, memories, model settings, or workspace files. Choose the provider you want to use and adjust the instructions for your project. Check its routines too: ones marked active in the shared setup stay active when installed. ## Add a Plugin or Skill to an Agent A plugin connects an agent to an app or service. Open a plugin listing to see what it can do and the example tasks it supports. Choose an agent, then select **Install plugin**. If the plugin includes skills, those are added to the agent you chose. The app connection itself is saved on the computer running OpenBot. Some plugins also need you to sign in or provide a credential before they can work. ![The GitHub plugin listing, with example requests, a selected agent, and the Install plugin button.](https://openbot.run/assets/github-plugin-ZXa_WYOF.png) The GitHub plugin listing shows example tasks and which agent will receive its skills. A skill is a set of instructions for a particular kind of task. In the **Skills** tab, open a listing, choose an agent, and select **Install skill**. A skill guides how the agent works; a plugin gives it a connection to another service. Check what access a plugin needs before installing it. The GitHub plugin, for example, can work with the repositories your personal access token can reach. Its token is saved on your computer, but it is sent to GitHub’s remote service with requests. Give it only the access you are comfortable sharing. ## Submit Your Agent to the Marketplace To share an agent, open the three-dot menu in the Marketplace and select **Add agent**. Before submitting, check the agent’s full instructions, skills, and routines in its settings. Remove any credentials or private details from that setup, and make sure its routines are suitable for someone else to run. ![The Marketplace menu with Discover, My submissions, Add agent, and Refresh options.](https://openbot.run/assets/marketplace-menu-GSrGeLgW.png) Use Add agent in the Marketplace menu to start a submission. Choose the local agent you want to share, review its name and description, and pick a category. The submission preview shows a summary and counts of included skills and routines; it does not show their full contents. Select **Submit for review** when you are ready. Conversations, memories, model settings, and workspace files are not part of the submitted template. You can follow the review in **My submissions**. Once an agent is approved, other people can find and install it from the Marketplace. If you need to revise a submission, you can send a new version. --- Source: https://openbot.run/guides/what-are-ai-agents [All guides](https://openbot.run/guides) # What Are AI Agents? How They Work and When to Use Them What are AI agents, how do they work, and when are they useful? A practical guide to their tools, use cases, and limits. Sep 24, 2026 Norbert Bodziony AI agents are software systems that use an AI model to work toward a goal. Depending on how they are built, they can take several steps, use connected tools, and adjust what they do based on what happens. This guide explains how AI agents work, what they are useful for, and when a chatbot or regular automation may be enough. ## What Is an AI Agent? The term is used in different ways, but an AI agent generally uses a model to carry out a task through one or more steps. The model generates responses and, in some setups, helps decide what to do next. Instructions set the agent's role and limits, context gives it information about the task, and connected tools let it interact with files or apps. Not every agent has the same setup. Some follow a mostly fixed process; others decide which steps to take as they go. Features such as persistent memory, a dedicated workspace, or a team of agents depend on the product—they are not requirements for every AI agent. ## How Do AI Agents Work? Say you ask an agent to compare three project-management tools for a small team. If it has browser access, it could check each tool's pricing and integrations, put the details into a table, and link to the pages it used. If a detail is missing, it should flag it rather than fill the gap with a guess. The agent uses the results of each step to decide whether to continue, try another approach, finish, or ask you for help. The tools it has been given determine what it can actually see and do; an agent does not automatically have access to your files or apps. ## AI Agents vs. Chatbots and Automation A chatbot describes how you interact with a system: through a conversation. An agent describes what the system can do. A chat interface can be the way you give instructions to an agent. A traditional automation follows steps set in advance. An agent can choose or adjust its next step based on the task and what it finds. Many systems combine the two. The terminology varies, but [Anthropic describes agents](https://www.anthropic.com/engineering/building-effective-agents) as systems that “dynamically direct their own processes and tool usage,” in contrast to workflows that follow predefined paths. ## What Are AI Agents Used For? With the right tools and permissions, an agent can help with work that involves several steps. For example, it could: - compare vendor plans using their pricing pages, then list the differences with source links; - read a code issue, inspect the relevant files, make a change, and report which tests it ran; - gather weekly figures from connected apps and prepare an update for review. What an agent can do depends on its model, tools, instructions, and permissions. ## When Should You Use an AI Agent? An agent is useful when you know the outcome you want, but the steps may depend on what happens along the way. Research, code changes, and reports based on several sources can all involve decisions during the task. For a quick question, a chatbot may be enough. If the same predictable steps need to run every time, a regular automation may be a better fit. An agent is most useful when it can move work forward across several steps without needing a new instruction after each one. ## What Are the Limitations of AI Agents? An agent can misunderstand a request, use incomplete or outdated information, or make a poor choice about what to do next. Errors can also build up over a longer task, so check important facts, sources, and changes before relying on the result. [Anthropic also notes](https://www.anthropic.com/engineering/building-effective-agents) that greater autonomy can bring higher costs and compounding errors in some systems. Access matters too. An agent can only work with the files, apps, and actions it has been allowed to use. For actions such as sending a message, changing a record, or publishing something, set clear limits and require approval when needed. ## How to Get Useful Results from an AI Agent Give the agent a clear outcome, the context it needs, and a way to tell when the task is done. Specify which tools it can use and which actions need your approval. For research tasks, ask it to link its sources and flag anything it could not verify. For example, instead of asking for a general comparison, ask for a table based on three official pricing pages, with links and missing details clearly marked. Review the result before letting the agent take any consequential action. ## AI Agents in OpenBot OpenBot gives each agent a role, workspace, and conversation. Agents can hand work and files to one another while the project stays together. OpenBot organizes the work around the connected provider; that provider still handles the model requests. OpenBot is local-first, but that does not necessarily mean the model runs on your computer. The workspace can be local while a connected provider handles the requests. For a closer look, see [How to Use OpenBot](https://openbot.run/guides/openbot-101). To see how this differs from an agent app that runs in the cloud, read [OpenBot vs Grok Bot](https://openbot.run/compare/grok-bot). --- Source: https://openbot.run/guides/wtf-is-openbot [All guides](https://openbot.run/guides) # What Is OpenBot? Models, Providers and Agents Explained A practical explanation of models, providers, agents, and the local-first workspace that brings them together. Sep 15, 2026 Norbert Bodziony At first glance, OpenBot might look like another app for chatting with AI. But it is not a model or a chatbot on its own. It is a workspace that brings the models and agents you already use into one place, together with their files, conversations, and context. To understand how it works, it helps to separate those pieces first. ## OpenBot Is Not a Model OpenBot does not create its own AI model. It also does not replace tools like Codex, Claude Code, Grok CLI, OpenCode, or models that you host yourself. The model is the part that generates the response. Its output depends on its training, the instructions it receives, the context around the task, and the tools or files it can access. Codex, Claude Code, Grok CLI, OpenCode, and similar tools act as provider integrations or runtimes. They connect OpenBot to a model and handle the communication, authentication, and provider-specific parts of the process. OpenBot is the workspace around that process. It gives agents a role, a workspace, a conversation, a queue, and a way to keep context, exchange files, and work with other agents. ## Model, Provider, Agent: What’s the Difference? A model generates the response. A provider integration or runtime connects that model to OpenBot. It can be Codex, Claude Code, Grok CLI, OpenCode, or another configured runtime. An agent is the persistent setup built around that connection. It has its own role, instructions, workspace, conversation, queue, and context. This means the model and the agent do not have to be the same thing. You can move an agent from one provider integration to another while keeping the work around it in place. For example, a research agent might use Claude for one task and Codex for another. The model may change, but the agent’s role, files, and conversation can remain the same. That is what makes an agent different from a saved prompt. It carries the work around the task, not just the instructions for answering it. ## What an Agent Actually Owns An agent in OpenBot is more than a name attached to a model. It is a persistent setup for a particular kind of work. It has its own identity, instructions, workspace, conversation, queue, and provider session. The workspace is where it works with files, while the conversation keeps a record of what has already happened. Even when you change the provider, the agent can keep working with the same files, conversation, and context. This setup becomes more useful when the work continues over time or has to go through several stages. ## Where the Work Happens OpenBot runs on your computer, so an agent’s workspace is a local folder rather than a chat that disappears when you close a browser tab. Its conversations, queues, attachments, and other app data stay with the desktop app and the machine that hosts it. That does not mean every part of the process stays local. When an agent uses Codex, Claude, Grok, or another provider, prompts, files, and tool results may be sent to that provider. The embedded browser also uses the network. Local-first describes where OpenBot keeps your project and agent state. It does not mean that the entire setup has to work offline. You can also keep OpenBot running on one computer, such as a Mac mini, and connect to it from other devices. The host remains the place where the agents and their workspaces live, while the other devices give you a way to follow the work. ## A Task From Start to Finish Imagine asking a research agent to prepare a short report. You give it the task in its conversation, and OpenBot runs it with the agent’s instructions, workspace, context, and selected provider. The agent can inspect local files, use the embedded browser, save notes, and return the result in the same thread. When the first part is done, you can pass the work to another agent. A writing agent can use the research files, while a review agent checks the result and points out what needs to change. Each agent can keep its own role and conversation, while the project stays in one place. You can follow the work as it happens, send another instruction, pause a task, or redirect it when the plan changes. The models still do the actual work. OpenBot keeps the tasks, files, conversations, and handoffs connected. ## Why Use More Than One Agent? One agent can handle a lot of work, but one setup is not always the best fit for every task. Research, coding, writing, and review often need different instructions, tools, and ways of working. Instead of asking one assistant to change roles constantly, you can give each agent a more focused job. OpenBot lets agents exchange messages and files, pass work between their conversations, and work together in shared channels. A research agent can gather information, a writing agent can turn it into a draft, and a review agent can check the result. More agents do not automatically make a task better. For a quick question, one agent may be enough. A larger setup becomes useful when the work has clear stages, needs different perspectives, or has to continue over time. ## Who Is OpenBot Actually Built For? OpenBot is for people who want AI work to continue beyond a single answer. It makes the most sense when a task involves files, ongoing context, repeated steps, or more than one role. A developer might keep a coding agent next to a research or review agent. A content team might use one agent for research, another for drafts, and another for feedback. An individual might keep an agent running recurring tasks instead of rebuilding the same setup every time. If you only need a quick answer, a regular chatbot may be faster. OpenBot becomes useful when the work starts to look more like a project than a conversation. It is also worth remembering that agents work with full access. They can access files, run commands, use the network, and control the browser, so you should only give them tasks you trust and keep backups of important work. ## More Than a Chat Window OpenBot is not a model and not just another chat app. It is a local-first workspace that keeps agents, models, files, conversations, and tasks connected. The model provides the responses. The provider integration connects it to OpenBot. The agent keeps the work around the task. OpenBot brings these pieces together so you can keep working instead of starting from a blank chat each time. That is the idea behind OpenBot: turning separate AI tools into persistent teammates for real work. If you want to see how it works in practice, start with [How to Use OpenBot](https://openbot.run/guides/openbot-101). To see how it differs from an agent app that runs in the cloud, read [OpenBot vs Grok Bot](https://openbot.run/compare/grok-bot). --- Source: https://openbot.run/guides/openbot-101 [All guides](https://openbot.run/guides) # How to Use OpenBot: Set Up Your First AI Agent What OpenBot is, what an agent keeps between runs, and how to get one doing real work on your computer. Start here if you have not opened the app yet. Sep 12, 2026 Norbert Bodziony I built OpenBot because I wanted teammates, not chats. A chat window forgets. A teammate keeps the folder it works in, the history of what it did, and a name you can call it by. Tomorrow morning it is still those things, on a different model, after a restart. This is the page I would send someone who has never opened the app. It covers what OpenBot is, what an agent actually owns, and how to get one doing work you can check. ## What OpenBot is OpenBot is a desktop workspace for AI teammates that live on your computer. It runs the coding command-line tools you already use as processes on that machine: [the Codex App Server](https://learn.chatgpt.com/docs/app-server), [Claude Code](https://code.claude.com/docs/en/overview), Grok CLI and OpenCode. Each agent gets a workspace, a thread and an identity around them. Your work sits in a SQLite database called `openbot.db`, in the application's data folder. That file is the source of truth. Workspaces, conversations, attachments, browser data and team data stay on the computer that runs OpenBot. **Local-first is not the same as offline.** Codex still talks to OpenAI, Claude to Anthropic, Grok to xAI. Pages in the embedded browser use the network. A plugin can call its own service. What does not leave is the record of the work. ## Anatomy of an agent Four things belong to an agent. All four survive a restart and a change of model: - **A workspace.** One directory of its own at `~/OpenBot/Agents/`. It also gets `~/OpenBot/Shared`, which is where agents hand files to each other. - **A thread.** Everything you have asked the agent, and everything it did, in one durable record. - **An identity.** A name, an avatar, a short role and the instructions you wrote for it. - **A private provider session.** The resume state of the command-line tool behind it. That one belongs to the provider, so it is kept apart from the thread on purpose. That is why a model is just a setting. Move an agent from Codex to Claude and the workspace, the thread and the name stay put. Only the thinking changes. ![An agent thread. Chief answers a launch question with a table of workstreams, owners and status, links the two working documents it read, and ends with a shell block.](https://openbot.run/assets/a-thread-C4YG4M8v.webp) A thread is the answer and the work behind it: mentions of other agents, the files that were read, and the command it wants run next. ## Install it, then connect a provider OpenBot runs on macOS 13 or newer on Apple silicon or Intel, Windows 10 or newer on x64, and x64 or arm64 Linux as an AppImage. Grab the installer from [GitHub Releases](https://github.com/nightly-labs/openbot/releases). The Windows installer is code-signed, but SmartScreen can still warn about a new release. Check the release checksum before you run it. On Ubuntu 23.10 or newer and on Debian 13, the AppImage needs an AppArmor profile first. [The README](https://github.com/nightly-labs/openbot#readme) has the two commands that install it. Then pick a provider. In onboarding, in Settings or in the model picker, select one and press **Download**. OpenBot installs and pins its own managed copy of that command-line tool. If you already installed the tool yourself and there is no managed copy, OpenBot uses yours. **OpenBot uses the login the command-line tool already has.** It does not copy your provider credentials. If `codex` or `claude` works in your terminal, it works here. ![The model picker open on the ChatGPT tab, showing the installed Codex CLI version, a list of GPT models, and a reasoning effort control.](https://openbot.run/assets/pick-a-model-BWAQz_tq.webp) The picker names the command-line tool and its version, so you can see which binary an answer came from. ## Your first task Make the first one small, specific and checkable. “Improve my project” gets you an essay. “Read the failing test in this file and tell me which claim it disproves” gets you something you can agree or disagree with. Then look at the activity under the answer, not only the answer. The thread shows the files the agent opened and the commands it ran. That is how you tell whether it did the work or described it. ## Full access, and what that means After you consent once at first launch, agents run with `danger-full-access` and `approvalPolicy: never`. They can read and change files, run commands, use the network and drive the embedded browser without asking again for each action. **This is a product decision, not a security boundary.** An agent starts in its own workspace, but these provider modes are unrestricted on purpose. They can reach outside that workspace wherever the operating system allows. Run agents and tasks you trust, and keep backups. What the application sends, and what it never sends, is written down in [the privacy document](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md). ## Channels: more than one agent on one thing A channel is a shared thread that several agents read and write. One agent owns the task at a time, and it hands the task on explicitly. You keep the controls: Stop, Resume, Reassign, Archive and Restore. It works best when the agents are different from each other. One drafts, one verifies, one owns the release date. Three copies of the same agent give you three copies of the same paragraph. ![A channel called Launch room. Chief assigns the release note to Launch, Research reports which performance claims it could verify, and Launch posts the finished draft as a linked file.](https://openbot.run/assets/a-channel-Cfnf--ZC.webp) One room, one task owner. The handover is a message, so you can read who took the work and why. ## Routines: the work that repeats A routine is one instruction plus the times to run it. The agent does the work in its own thread, so the result lands where the context already is, and the run history sits beside it. Size the instruction so that “nothing to report” is a normal answer. A routine that must always find something will always invent something. ![A routine in a thread. Two invoked routine markers sit above their instructions, and a side panel shows the schedule, on weekdays at 9:00 AM, with a successful run yesterday.](https://openbot.run/assets/a-routine-CV81WGxY.webp) The thread marks which turns a routine started, so a scheduled answer never reads as one you asked for. ## Your team, on your own computer You can invite people to your OpenBot instead of copying it. Publishing never starts a second instance. The Team API stays on loopback, and a hidden sandboxed page connects invited clients over WebRTC. Signal only carries the connection setup. Cloudflare holds accounts, avatars, host configuration, memberships, invitations and logical session records. It does not carry chats, files or commands. ## What is next [The news section](https://openbot.run/news) is where I write about the decisions behind these parts. [The guides section](https://openbot.run/guides) is where the how-to pages collect. The [repository](https://github.com/nightly-labs/openbot) holds the code, and [the releases page](https://github.com/nightly-labs/openbot/releases) holds every build. If something here is wrong, or a step does not match what you see, tell me. A preview is easier to correct than a release. --- Source: https://openbot.run/compare/paperclip [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Paperclip: An App, Not a Server to Set Up OpenBot vs Paperclip: a desktop app with iPhone and Android apps and no VPN, or a Node.js server with an org chart and budgets. Setup, phones, price, license. Checked Oct 23, 2026 Sources ## The short answer Choose OpenBot if you want an app that you download and open, agents that you chat with as a team, and iPhone and Android apps that reach them from anywhere with no VPN: the app is free. Choose Paperclip if you want to run agents as a company, with an org chart, goals, budgets and approval stages, on a Node.js server that you set up, under the MIT license. ### Choose OpenBot if Recommended - You want a desktop app for macOS, Windows or Linux, with no server, Node.js or database to set up. - You want iPhone and Android apps that reach your agents from anywhere, with no Tailscale, VPN or public address. - You want to chat with your agents, and see them give work to each other in shared channels. - You do not want to keep a computer on, and want a hosted server that is available today. ### Choose Paperclip if - You want an org chart, goals, budgets for each agent and approval stages, as in a company. - You need the MIT license, for example for commercial use. - You run several companies, each with its own agents and data, in one deployment. - You want to connect many agent tools, or your own scripts and HTTP services, as agents. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Paperclip:** Paperclip runs the agent tools that you install and sign in yourself, such as Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi, Hermes Agent, Grok Build, Kimi Code and OpenClaw, or any process or HTTP service that can get a heartbeat. Paperclip itself is free under the MIT license. ## At a glance OpenBot is better on 7 of 13 topics, and Paperclip on 1. OpenBot and Paperclip compared, as checked on Oct 23, 2026 | Topic | OpenBot | Paperclip | | --- | --- | --- | | Models and plans | Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Bring your own agent tools: Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi, Hermes Agent, Grok Build, Kimi Code, OpenClaw, or a script or HTTP service. You install and sign in to each one. | | Where agents work | On the computer that runs OpenBot. | On the computer that runs the Paperclip server, or in remote sandboxes and cloud agents that you connect to it. The Paperclip server runs on your laptop, a server, Docker, or a cloud host that you set up. | | How you work with agents | Better: You chat with them. Agents give work to each other in shared channels, and you step in when a decision needs you. They can also answer in Slack, Discord and Telegram. | Through tickets on a dashboard. Agents wake on a schedule or when a task comes to them. Agent Chat and chat connectors are experimental, and you turn them on in the instance settings. | | When your computer is off | Better: Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | The Paperclip server must stay on to start its agents. Paperclip Cloud has a waitlist only, with no date or price; until it opens, you set up and run your own server. | | Phones and remote access | Better: Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | No app for iPhone or Android: you open the web UI in a browser. By default it listens only on the computer itself; to reach it from your phone, you set up Tailscale, a VPN, your local network or a public address. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | You choose the agent tool and the model for each agent. When a Claude plan reaches its limit, the agent waits until the limit resets. It does not move to a different model or provider. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | Paperclip keeps API keys and tokens encrypted on the computer that runs it, and gives them to the agent when it starts. The agent can read each secret that it gets. Paperclip has no password manager and does not fill in logins or payment cards. | | Voice | Better: Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | The documentation states no voice mode. The code has an experimental voice connection through Speko, a hosted service that handles the audio. | | Teams | Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | Agents have a boss, a title and a job in an org chart, with budgets and approval stages. People can join after you turn on sign-in from the server's shell. | | Your data | Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | By default, tasks, agents and files stay in a Postgres database and in files on the computer that runs Paperclip. Usage telemetry, with no prompts or model output, is on by default; you can turn it off. | | Price and account | Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Better: Free under the MIT license, also for commercial use, with no Paperclip account. You pay for the agent tools, and for a server if you rent one. | | Apps and setup | Better: Download the app for macOS, Windows or Linux and open it. Mobile apps for iPhone and Android. | A Node.js server with a web UI. You install Node.js 24.11 or later and start it from a terminal with npx. | | Source code | On GitHub, under the PolyForm Noncommercial License 1.0.0. | On GitHub, under the MIT License. | Paperclip says: "If OpenClaw is an employee, Paperclip is the company." It is a Node.js server with a web UI that runs agent tools such as Claude Code and Codex as a company: an org chart, goals, tickets, budgets and approval stages. OpenBot also runs a team of agents, on a computer that you choose, with the AI plans you already pay for. The difference is how you set it up and how you work with the agents. Paperclip is a server that you start from a terminal and reach in a browser, and you manage tickets. OpenBot is an app that you download. You chat with your agents, and your phone reaches them from anywhere. ## Models and the plans you pay for ### OpenBot OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Paperclip Paperclip does not run models itself: it runs agent tools through adapters. It lists Claude Code, Codex, Cursor and Cursor Cloud, Gemini CLI, OpenCode, Pi, Hermes Agent, Grok Build, Kimi Code and OpenClaw, and also runs your own processes, HTTP services and adapter packages. Paperclip installs no agent tools for you, unless you use its Docker or sandbox image: you install each one and sign in with its own account or API key. You choose the tool and the model for each agent. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Paperclip Paperclip runs as one Node.js process on the computer that you choose, with an embedded Postgres database and files on that computer. You can also use your own Postgres, run it in Docker, or deploy it to a server or a cloud host. Agents can also work in remote sandboxes, such as Daytona or e2b, but the server must stay on to start them. Agents wake on a schedule or when a task comes to them. Paperclip Cloud has a waitlist, with no published date or price. ## Setup ### OpenBot Better Download OpenBot for macOS, Windows or Linux, open it, and sign in to the provider tools that you want to use. OpenBot downloads and pins its own copy of each provider tool, or uses the one you installed. You set up no server, runtime or database: the app keeps its data in a SQLite database on your computer. ### Paperclip You install Node.js 24.11 or later and run npx paperclipai onboard in a terminal. Paperclip then runs as a local server, and you open its web UI in a browser. By default it is in local trusted mode: no sign-in, and it listens only on the computer itself. For other people or devices, you change the deployment mode on the server and claim ownership with a one-time link from the server log. ## Phones and remote access ### OpenBot Better The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Paperclip Paperclip says that you can manage your agents from your phone, in its web UI; it lists no app for iPhone or Android. To reach the server from another device, you switch it to authenticated mode and then use Tailscale, a VPN or your local network, or a public deployment with its own public URL. ## How agents work as a team ### OpenBot In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You chat with them, follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Paperclip Paperclip organises agents as a company. Each agent has a boss, a title and a job in an org chart, and work is in tickets that trace back to the company goals. Agents wake on a schedule or when a task comes to them, and give work up and down the org chart. You approve new agents and set a budget for each one; at the limit, Paperclip pauses the agent. One deployment can run several companies, each with its own agents and data. Agent Chat is experimental. ## Data and privacy ### OpenBot OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Paperclip Paperclip keeps tasks, agents, comments and files in a Postgres database and in files on the computer that runs it, and needs no Paperclip account. Usage telemetry is on by default; its rules exclude prompts, model output, secrets and private paths. You turn it off with PAPERCLIP_TELEMETRY_DISABLED=1, DO_NOT_TRACK=1, or its config file. ## Questions ### Is OpenBot an alternative to Paperclip? Yes. Both run a team of AI agents on a computer that you choose, with agent tools such as Codex and Claude Code. OpenBot is an app that you download. You chat with your agents, and your phone reaches them from anywhere. Paperclip is a server with a web UI, and you manage your agents through tickets, an org chart and budgets. ### What is the main difference between OpenBot and Paperclip? How you work with the agents. In OpenBot, you chat with a team of agents, and they give work to each other in shared channels. In Paperclip, you run the agents as a company: you set goals, approve new agents and budgets, and follow their work in tickets. ### Does Paperclip have an app for iPhone or Android? Paperclip lists no app for iPhone or Android. Its web UI works in a phone's browser, when your phone can reach the server. OpenBot has apps for iPhone and Android that connect from anywhere. ### Do I need Tailscale or a VPN to reach my agents from my phone? With OpenBot, no: the apps connect over an encrypted connection between your devices, with no VPN or tunnel. With Paperclip, the server listens only on its own computer by default; you switch it to authenticated mode, and reach it through Tailscale, a VPN, your local network or a public deployment. ### Is there a Paperclip cloud? Paperclip Cloud has a waitlist, with no published date or price. Until then, you run Paperclip on your own computer or server. If you do not want to keep a computer on for OpenBot, a hosted OpenBot server in the EU is available today, from €30 or $36 a month. ### Do agents keep working when my laptop is closed? Yes, with both, when they run on a computer that stays on. With OpenBot, run it on a desktop, a server or a hosted OpenBot server, and connect from your laptop or phone. With Paperclip, run the server on a computer or a cloud host that stays on. ### How much do OpenBot and Paperclip cost? Both are free. OpenBot is free for noncommercial use, and commercial use needs a separate license. Paperclip is free under the MIT License. With both, you pay for the agent tools or the model provider, and for a server if you rent one. ### Is Paperclip open source? Yes. Paperclip is on GitHub under the MIT License, which also allows commercial use. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0, and commercial use needs a separate license. ## Sources Every statement about Paperclip comes from the pages below, checked on Oct 23, 2026. They can change after that date. - [Paperclip](https://paperclip.ing/) - [Paperclip on GitHub](https://github.com/paperclipai/paperclip) - [Paperclip license](https://github.com/paperclipai/paperclip/blob/master/LICENSE) - [Paperclip for AI assistants (llms.txt)](https://paperclip.ing/llms.txt) - [Paperclip documentation](https://docs.paperclip.ing/) - [Paperclip: enable multi-user login](https://docs.paperclip.ing/how-to/enable-multi-user-login/) - [Paperclip: deployment modes](https://github.com/paperclipai/paperclip/blob/master/docs/deploy/deployment-modes.md) - [Paperclip: Tailscale private access](https://github.com/paperclipai/paperclip/blob/master/docs/deploy/tailscale-private-access.md) - [Paperclip telemetry settings](https://github.com/paperclipai/paperclip/blob/master/packages/shared/src/telemetry/config.ts) - [Paperclip telemetry rules](https://github.com/paperclipai/paperclip/blob/master/packages/shared/src/telemetry/README.md) - [Paperclip Cloud waitlist](https://paperclip.ing/waitlist/) - [Paperclip Claude adapter](https://docs.paperclip.ing/adapters/claude-local) - [Paperclip secrets](https://docs.paperclip.ing/deploy/secrets) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/viktor [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Viktor: A Private, Local Alternative OpenBot is a free, private Viktor alternative: run a team of AI agents on your own computer, with the AI plans you already pay for, no credits, and your data kept there. Checked Oct 23, 2026 Sources ## The short answer Choose OpenBot if you want a team of AI agents on your own computer, with your data in a database that you hold and the AI plans you already pay for: the app is free, and there are no credits to buy. Choose Viktor if you want one AI coworker in Slack or Microsoft Teams that runs in the cloud, connects to thousands of business apps, and comes with a SOC 2 report. ### Choose OpenBot if Recommended - Your chats, files and company data must stay on your own computer, not on a service's servers in the US. - You already pay for ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code, and you do not want to pay for each task again in credits. - You want a team of agents, each with its own job and model, not one assistant for the whole company. - You want a free app that you can download today, with source code that you can read. ### Choose Viktor if - Your team works in Slack or Microsoft Teams, and you want one AI coworker that everyone mentions there. - You need a ready connection to many business apps, such as Salesforce, HubSpot and Google Ads. - Your company needs a vendor with a SOC 2 Type 1 report, SAML single sign-on and a DPA. - You want no computer of yours to stay on. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Viktor:** You choose from the models that Viktor offers, from OpenAI, Anthropic, Google and xAI, or connect your own OpenRouter key. An admin sets the default model for the workspace. You pay in credits: up to $100 in free credits to start, then a Team plan from $50 a month for 20,000 credits. ## At a glance OpenBot is better on 9 of 13 topics, and Viktor on 1. OpenBot and Viktor compared, as checked on Oct 23, 2026 | Topic | OpenBot | Viktor | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Models from OpenAI, Anthropic, Google and xAI, or your own OpenRouter key. You cannot use a ChatGPT or Claude plan that you already pay for, and Viktor bills each task in credits. | | Price | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for, so there is no charge for each task. | Up to $100 in free credits, then from $50 a month for 20,000 credits. Viktor says that 40,000 credits are about 80 to 400 tasks, and scheduled tasks use credits too. | | Where agents work | Better: On the computer that runs OpenBot, or on a hosted OpenBot server in the EU. | On Viktor's own cloud computer for your workspace, in the US. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | Stored in the US. Viktor can read all messages in the Slack channels it joins, not only those that mention it, and can get up to about 90 days of earlier history. EU data residency is only on Enterprise contracts. | | Teams | Better: Each agent is a full coding agent with its own job, model and workspace. They give work to each other in shared channels, and coworkers can join one team host. Roles set which agents each member can view, chat with or manage. | One shared AI coworker for the whole workspace, with one pool of credits. | | Who approves an action | You do. An agent shows a draft and waits for your approval before it sends a message for you, unless you or a routine that you set up tells it to send. It asks you to confirm the item, price and total before a payment. | An admin chooses which tools need approval. Under Viktor's terms, you are responsible for every message and action that it sends or takes for you. | | Integrations | Agents use MCP servers, plugins from the OpenBot catalog, and the command-line tools on your computer. | Better: More than 9,800 apps, such as Salesforce, HubSpot, Linear, Notion, Stripe and Google Ads. | | How you talk to it | The desktop app, and apps for iPhone and Android over an encrypted connection between your devices. Agents can also answer in Slack, Discord and Telegram. | In Slack, Microsoft Teams or Discord, or in the Viktor web app. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | An admin chooses the default, fast and deep models. Viktor states a backup only to a second cloud host for Claude Fable 5. When credits run out, Viktor pauses. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | Viktor keeps the OAuth tokens and API keys of your apps in encrypted vaults and adds them on its servers, so the model does not see them. It states no saved passwords, password manager or payment cards for its browser. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | You can talk to Viktor in a live voice call, and it speaks its replies. A setting stops it when you talk over it. ElevenLabs and Groq process the speech in the cloud. | | Who can use it | Better: Anyone, in any country: OpenBot has no region lock. Each AI provider sets the countries for its own plan. The desktop app is in 15 languages, and hosted servers run in the EU. | Businesses only: its terms say that the service is not for consumers. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | A closed service that runs only on Viktor's servers. | Viktor is an AI coworker that lives in Slack and Microsoft Teams. Anyone on the team mentions it, and it does the work on its own cloud computer: reports, ad campaigns, bug triage, internal apps and invoices, with more than 9,800 connected apps. OpenBot also gives you AI agents that do real work. The difference is where they run, who holds your data, and how you pay. Viktor runs in the US, can read the channels it joins, and bills each task in credits. OpenBot runs a team of agents on your own computer, keeps your data there, and uses the AI plans you already pay for. ## Models and what you pay ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. OpenBot is free for noncommercial use, so a busy month does not cost more. ### Viktor Viktor offers models from OpenAI, Anthropic, Google and xAI, and an admin sets the default model for the workspace. You can also connect your own OpenRouter key; those models are billed to your OpenRouter account. You cannot use a ChatGPT or Claude plan that you already pay for. You pay in credits, shared by the whole workspace. You start with up to $100 in free credits. The Team plan starts at $50 a month for 20,000 credits and goes up to $50,000 a month for 20 million credits; Enterprise has custom prices. Viktor says that a quick Slack question uses about 50 to 100 credits, a deep research report 500 to 2,000, and a full project 2,000 to 5,000. Scheduled tasks use credits each time they run, and unused credits roll over for one month only. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Viktor Viktor has its own computer in the cloud, where it writes and runs code. Its security page says that it is hosted on AWS in us-east-1, and that each workspace runs in a sandbox. No computer of yours must stay on. Data is in the US by default; EU data residency is only on Enterprise contracts. ## What it reads, and who holds your data ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Viktor Viktor's privacy policy says that it reads all messages in the Slack channels it joins, not only the messages that mention it, and that it can get up to about 90 days of earlier history when it joins. A user can give it wider search access, which can include private channels and direct messages. Data is stored with cloud providers in US regions, and the policy names more than 30 other services that process it. Viktor says that it does not train foundation models on your data, but it can use your data for small internal models, such as one that chooses the model for a task. ## How agents work as a team ### OpenBot Better In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job, model and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. When you move an agent to a different provider, it keeps its role, workspace and conversation. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Viktor Viktor is one AI coworker for the whole workspace. Everyone on the team mentions the same Viktor, it keeps one shared memory of the team, and every request uses the same pool of credits. ## Approvals and responsibility ### OpenBot OpenBot agents show a draft before they send a message for you. They send it after you approve it, or when you or a routine that you set up tells them to send. Before a payment, an agent reads back the item, price and total and asks you to confirm. A new agent has Full access: it does not ask before it runs a command. Set an agent to Ask first, and that agent asks you first; you answer in the chat or from a notification on your phone. ### Viktor Some built-in actions, such as joining a Slack channel, always need approval in Viktor. For each integration tool, an admin chooses if it runs on its own, needs approval, or never runs, and a user who can edit that tool can click Always approve (not in Microsoft Teams). Viktor's terms make you responsible for all messages and actions sent or taken through your account, and you must defend Viktor against third-party claims that come from any action it takes through a connected app under the permissions you set. ## Integrations and company controls ### OpenBot OpenBot agents use MCP servers, plugins from the OpenBot catalog, and the tools on the computer that runs them. They can also answer in Slack, Discord and Telegram. OpenBot has no SOC 2 report. On your own computer, your own security controls apply to your data. ### Viktor Better Viktor connects to more than 9,800 apps and builds a custom integration for one it does not have. It has a SOC 2 Type 1 report; SOC 2 Type 2 and ISO 27001 are in progress. Enterprise adds SAML single sign-on, role-based access, audit logs and a DPA. ## Questions ### Is OpenBot an alternative to Viktor? Yes, if you want your agents and your data on your own computer. Both give you AI agents that do work for your team. OpenBot runs a team of agents on your computer, with the AI plans you already pay for. Viktor is one AI coworker in Slack or Microsoft Teams that runs in the US and bills each task in credits. ### How much does Viktor cost? Viktor gives up to $100 in free credits. Then the Team plan starts at $50 a month for 20,000 credits; Viktor says that 40,000 credits are about 80 to 400 tasks. Enterprise has custom prices. OpenBot is free for noncommercial use, and it uses the AI plans you already pay for, with no credits. ### Can I use my ChatGPT or Claude plan with Viktor? No. Viktor bills its models in credits, or you connect your own OpenRouter key. OpenBot runs Codex with your ChatGPT plan, Claude Code with your Claude plan, and Gemini, Grok, Cursor, Cline and Muse Code with your own accounts. ### Does Viktor read all my Slack messages? Viktor's privacy policy says that it reads all messages in the channels it joins, not only those that mention it, and up to about 90 days of earlier history. A user can give it wider access, which can include private channels and direct messages. OpenBot keeps your workspaces, chats and files in a database on your own computer. ### Where does Viktor store my data? In the US: Viktor is hosted on AWS in us-east-1, and EU data residency is only on Enterprise contracts. OpenBot keeps your data on your computer, or on a hosted OpenBot server in the EU that you choose. ### Do agents keep working when my laptop is closed? Yes, with both. Viktor works on its own cloud computer, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### Is Viktor open source? No. Viktor is a closed service. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0. ## Sources Every statement about Viktor comes from the pages below, checked on Oct 23, 2026. They can change after that date. - [Viktor](https://viktor.com/) - [Viktor pricing](https://viktor.com/pricing) - [Viktor security](https://viktor.com/security) - [Viktor changelog](https://viktor.com/changelog) - [Viktor approvals](https://viktor.com/docs/approvals) - [Viktor integration permissions](https://viktor.com/docs/integration-permissions) - [Viktor chat commands](https://viktor.com/docs/chat-commands) - [Viktor enterprise](https://viktor.com/enterprise) - [Viktor privacy policy](https://viktor.com/legal/privacy) - [Viktor general terms](https://viktor.com/legal/general-terms) - [Viktor settings](https://viktor.com/docs/settings-overview) - [Viktor: running out of credits](https://viktor.com/docs/running-out-of-credits) - [Viktor: what Viktor can do](https://viktor.com/docs/what-viktor-can-do) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/instinct [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Instinct: A Private, Local Alternative OpenBot is a free, private Instinct alternative: run a team of AI agents on your own computer, keep your data there, and approve each message before it is sent. Checked Oct 15, 2026 Sources ## The short answer Choose OpenBot if you want a team of AI agents on your own computer, with your data in a database that you hold, the AI plans you already pay for, and agents that show you a draft before they send a message for you: the app is free. Choose Instinct if you have an invite, and you want one assistant that you text or call, with no app to set up. ### Choose OpenBot if Recommended - Your mail, files and chats must stay on your own computer, not on a service's servers in the US. - You want to approve a message or a payment before an agent makes it. - You want ChatGPT, Claude, Gemini, Grok, Cursor, Cline, Muse Code and Pi agents in one team, or your own model. - You want a free app that you can download today, with source code that you can read. ### Choose Instinct if - You have an invite, and you want an assistant with no app: you text it or call it. - You want it to use a phone and a computer for you, as a person does. - You want no computer of yours to stay on. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Instinct:** Instinct names no model. It is in private access, by invite, and it publishes no price; its terms say that it can charge fees and change its prices at any time. ## At a glance OpenBot is better on 10 of 13 topics, and Instinct on 0. OpenBot and Instinct compared, as checked on Oct 15, 2026 | Topic | OpenBot | Instinct | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Instinct names no model. Its privacy policy says that it shares data with third-party AI model providers. | | Where agents work | Better: On the computer that runs OpenBot. | On Instinct's service, which its terms say it operates in the United States. Instinct says that it can use a phone and a computer as people do. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Instinct works on its own service, so no computer of yours must stay on. | | How you talk to it | Apps for iPhone and Android connect from anywhere, over an encrypted connection between your devices, and no cloud stores your chats or files. Agents can also answer in Slack, Discord and Telegram. | By text message, WhatsApp or a phone call. Instinct has no app of its own to talk to it. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Instinct names no model and states no backup model. You cannot choose a model. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | You can give Instinct your passwords for other accounts and your payment details, so that it signs in and books for you. Its privacy policy names no encryption and says that it cannot guarantee perfect security. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | Instinct can use your data to train its AI models, unless you opt out. A disconnected app does not delete the data that Instinct collected from it, and when you delete your account, Instinct does not have to delete your materials. | | Who approves an action | Better: You do. An agent shows a draft and waits for your approval before it sends a message for you, unless you or a routine that you set up tells it to send. It asks you to confirm the item, price and total before a payment. | Under its terms, you appoint Instinct as your agent to make agreements and transactions for you, and they bind you as if you had made them. One tester said that it sent an email for her without checking first. | | Teams | Better: Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | One personal assistant for each person. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | No published price. The terms say that you pay the fees that apply, and that Instinct can change its prices at any time. You sign in with a phone number. | | Where you can use it | Better: In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | By invite only, in private access. Instinct operates its service in the United States and names no countries. Only for people 18 and over. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | No app to install: you text or call Instinct, and you manage your account on the web. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | A closed service that runs only on Instinct's servers. | Instinct is a personal assistant that you text or call. It connects to your email, messaging apps and calendar, and to the screen, audio and location of your devices, and it acts for you: it books tables, cleans up an inbox and shops. It is in private access, by invite. OpenBot also gives you agents that do work for you. The difference is where they run, who holds your data, and who approves what they do. Instinct runs on its own service in the US, can train its models on your data, and its terms make the transactions it enters bind you. OpenBot runs a team of agents on your own computer, keeps your data there, and asks you before it sends a message or pays. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Instinct Instinct does not say which models it uses, and you cannot choose one. Its privacy policy names third-party AI model providers among the parties that get your data. Instinct publishes no price: it is in private access, and its terms say that you pay the fees that apply and that it can change its prices at any time. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Instinct Instinct works on its own service, which its terms say it operates in the United States, so no computer of yours must stay on. It connects to your email, messaging apps and calendar, and to the screen, audio and location of your devices. Instinct says that it can use a phone and a computer in the same way that people do. ## Who approves what the agent does ### OpenBot Better OpenBot agents show a draft before they send a message for you. They send it after you approve it, or when you or a routine that you set up tells them to send. Before a payment, an agent reads back the item, price and total and asks you to confirm. A new agent has Full access: it does not ask before it runs a command. Set an agent to Ask first, and that agent asks you first; you answer in the chat or from a notification on your phone. ### Instinct Instinct's terms authorize it to take the actions that it decides answer your request, and appoint it as your agent to enter into agreements, commitments or transactions for you. Those bind you as if you had entered into them yourself. Testers told TechCrunch that Instinct sent an email for one of them without checking first, and found a sign-up code in another's inbox to complete a booking. ## Phones and remote access ### OpenBot The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Instinct You talk to Instinct by text message or WhatsApp, or you call it; there is no Instinct app to install. You sign in on the web with your phone number and a code, and you manage connected apps and your data there. ## How agents work as a team ### OpenBot Better In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Instinct Instinct is one personal assistant for each person. It describes no teams of agents and no shared workspace. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Instinct Instinct's privacy policy says that it can use your information to train the AI models behind its products. You can opt out, but models already trained on your data stay, and data flagged for safety review can still be used. It does not use data from Google Workspace to train models. Disconnecting an app does not delete the data that Instinct collected from it; you delete that data on the web. When you delete your account, Instinct may delete your materials, but it does not have to. A tester told TechCrunch that Instinct still summarized her inbox after she disconnected it, and that it said it kept the emails in plain text. ## Questions ### Is OpenBot an alternative to Instinct? Yes, if you want your agents and your data on your own computer. Both give you AI agents that do work for you. OpenBot runs a team of agents on your computer, with the AI plans you choose, and asks you before it sends a message or pays. Instinct is one assistant that you text or call, on its own service in the US. ### Is Instinct private? Instinct connects to your email, messages and calendar, and to the screen, audio and location of your devices. Its privacy policy says that it can train its AI models on your information unless you opt out, that it shares data with third-party AI model providers, and that disconnecting an app does not delete the data it collected. OpenBot keeps your workspaces, chats and files in a database on your own computer. ### Can Instinct act for me without asking? Its terms authorize it to take actions for you and to enter into agreements and transactions that bind you as if you had made them. A tester said that it sent an email for her without checking first. OpenBot agents show a draft and wait for your approval before they send a message, unless you tell them to send, and confirm before a payment. ### How much does Instinct cost? Instinct publishes no price. It is in private access, by invite, and its terms say that you pay the fees that apply and that it can change its prices at any time. OpenBot is free for noncommercial use, and it uses the AI plans you already pay for. ### Does Instinct run on my computer? No. Instinct runs on its own service, which it operates in the United States, and you reach it by text message, WhatsApp or a phone call. OpenBot agents work on your computer, or on a hosted OpenBot server in the EU that you choose. ### Do agents keep working when my laptop is closed? Yes, with both. Instinct works on its own service, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### Is Instinct open source? No. Instinct is a closed service. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0. ## Sources Every statement about Instinct comes from the pages below, checked on Oct 15, 2026. They can change after that date. - [Instinct](https://instinct.com/) - [Instinct terms of service](https://instinct.com/terms) - [Instinct privacy policy](https://instinct.com/privacy-policy) - [Instinct sign-in](https://app.instinct.com/) - [TechCrunch: Instinct's AI assistant raises privacy and security concerns](https://techcrunch.com/2026/08/24/instincts-powerful-ai-assistant-is-raising-privacy-and-security-concerns/) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/best-ai-agent-apps [All comparisons](https://openbot.run/compare) Roundup # Best AI Agent Apps in 2026: 13 Apps Compared 13 AI agent apps compared: OpenBot, Claude Code, Codex, Cursor, Antigravity, Devin, Manus and more. Where they run, models, price and benchmark scores. Checked Oct 3, 2026 Sources ## The short answer For one coding agent on one AI plan, pick Claude Code, Codex, Cursor or Antigravity. For an agent that keeps working with your computer off, pick a cloud agent like Devin, Manus, Claude Cowork or ChatGPT dots. If you want several of these agents working together on your own computer, with plans you already pay for, that's what we built OpenBot for. We make OpenBot, so it's first on the list. Every other entry links to a comparison with its sources. ## At a glance AI agent apps compared, as checked on Oct 3, 2026 | App | Best for | Runs on | Models | Price | | --- | --- | --- | --- | --- | | OpenBot | A team of agents on your own computer, with the AI plans you already pay for. | Your computer. Apps for macOS, Windows, Linux, iPhone and Android, and a web app. | Your ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code plan, free models, Inception's Mercury models, your own model, or any ACP agent. | Free for noncommercial use. Hosted EU server from €30 or $36 a month. | | Claude Code | Claude's models in your terminal and your IDE. | Your computer, or cloud sessions at Anthropic. | Claude models. | Claude Pro from $17 a month billed yearly. Max $100 or $200. | | Codex | OpenAI's models, with a sandbox that is on by default. | Your computer, or cloud tasks at OpenAI. | OpenAI models. | ChatGPT Plus $20 a month. Pro $100, $200 or $500. | | Cursor | A full code editor with agents, and models from many providers. | Your computer, cloud virtual machines, or machines that you host. | Grok, Composer, Claude, GPT, Gemini and Muse models. | Hobby is free. Pro $20 a month, Pro+ $60 and Ultra $200. | | Antigravity | Gemini models, with a free plan to start. | Your computer. Remote Control connects a browser to it. | Gemini models, with Claude Sonnet 5.5 and Opus 5.5 on Pro and Ultra. | Free with weekly limits. Google AI Pro $19.99 a month. | | Devin | An autonomous software engineer in the cloud. | Virtual machines in Cognition's cloud, or your computer. | Models from Anthropic, OpenAI, Google, Cognition and open-source labs. | Free with a light quota. Pro $20 a month, Max $200. | | OpenClaw | An open-source personal assistant in your messaging apps. | Your computer, a home server or a VPS that you rent. | Your ChatGPT, Claude Code, SuperGrok or Copilot login, API keys, or local models. | Free under the MIT License. | | Hermes Agent | An open-source agent that learns skills over time. | Your computer, a server, Docker, or Hermes Cloud. | API keys from many providers, local models, or a ChatGPT, SuperGrok or Copilot login. | Free under the MIT License. Hermes Cloud costs extra. | | Claude Cowork | Claude as an agent for knowledge work, in Anthropic's cloud. | Anthropic's cloud. | Claude models only. | A paid Claude plan: Pro from $17 a month, Max from $100. | | Manus | Research, websites and slides from one prompt. | A cloud sandbox that Manus runs, or folders on your computer that you allow. | Manus chooses the model for each task. | Free with 300 credits a day. Pro from $20 a month. | | ChatGPT dots | One always-on agent for ChatGPT Pro and Business Premium users. | A cloud computer that OpenAI runs, or your computer if you allow it. | GPT-6 Astra. | Included with ChatGPT Pro, from $100 a month. | | Grok Bot | Bots on a cloud computer, for Cursor and SuperGrok users. | A cloud computer that Cursor hosts, in the United States. | Cursor manages model selection. | Included with paid Cursor plans and SuperGrok. | | Muse | A personal assistant for email and purchases. | A cloud computer that Meta hosts. | Muse Spark, Meta's own model. | Free with a weekly limit. $20 or $100 a month. | ## By the numbers Four apps on this list are coding agents that Artificial Analysis tests. The best runs of Claude Code, Codex and Devin land between 62 and 68. Antigravity gets 64 only with a model that isn't public yet, and 42 with its default. Cost per task varies far more than score: from $1.04 for Codex to $14.19 for Claude Code. OpenBot isn't on the index. It runs these agents rather than competing with them. Artificial Analysis Coding Agent Index, read on Oct 3, 2026 | Agent and model | Index score Higher is better | API cost per task Lower is better | Time per task Lower is better | | --- | --- | --- | --- | | Claude Code Sonnet 5.5 (max) | 68 | $14.19 | 87 min | | Codex GPT-6.1 Sol (xhigh) | 63 | $1.04 | 16 min | | Devin Fusion CLI Claude Fable 5.1 XHigh + SWE-2 Medium | 62 | $7.90 | 36 min | | Antigravity CLI Gemini 4 Argon (high)Not public yet | 64 | $5.84 | 35 min | | Antigravity SDK Gemini 3.8 Flash (high)Tested with the SDK, not the CLI | 42 | $2.47 | 12 min | Data from the [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents), read on Oct 3, 2026. Cost is what a task costs at pay-per-token API prices, not a plan price. Each row is one agent with one model, tested the [way Artificial Analysis describes](https://artificialanalysis.ai/methodology/coding-agents-benchmarking). An AI agent app gives a model tools. It reads and writes files, runs commands, browses the web and keeps going until the task is done. The apps below differ in three ways: where the agent works (your computer or someone else's), which models it uses and who you pay for them, and how many agents can work together. After OpenBot, the list runs from coding agents to personal assistants. It's not a ranking. ## The apps 1. ### 1.OpenBot Best for: A team of agents on your own computer, with the AI plans you already pay for. OpenBot runs Codex, Claude Code, Antigravity, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code and custom ACP agents as one team on your computer. Each agent gets a job, and they pass work to each other in shared channels. When a plan hits its limit, an agent can move to a backup model on another provider. You follow along from the iPhone and Android apps or the web app, no VPN needed, and agents answer in Slack, Discord and Telegram too. Workspaces, chats and files stay in a database on your computer. 2. ### 2.Claude Code Best for: Claude's models in your terminal and your IDE. Anthropic's coding agent works in the terminal, VS Code, JetBrains IDEs, a desktop app and the web. Cloud sessions run in VMs that Anthropic manages, and you can follow them from the Claude phone app. It has the top score on the Artificial Analysis Coding Agent Index, 68, though that run costs $14.19 a task. It also works with an Anthropic API key, Bedrock, Google Cloud and Microsoft Foundry. - [Codex vs Claude Code: Which Coding Agent to Use](https://openbot.run/compare/codex-vs-claude-code) - [Claude Code vs Antigravity: Claude or Gemini Agent](https://openbot.run/compare/claude-code-vs-antigravity) - [Cursor vs Claude Code: Editor or Terminal Agent](https://openbot.run/compare/cursor-vs-claude-code) 3. ### 3.Codex Best for: OpenAI's models, with a sandbox that is on by default. OpenAI's coding agent works in a command line, the ChatGPT desktop app, and VS Code, Cursor and Windsurf. The sandbox is on from the first run, and the command line is open source under Apache-2.0. Cloud tasks run in VMs that OpenAI manages. It's the cheap one on the benchmark: 63 for $1.04 a task with GPT-6.1 Sol. - [Codex vs Claude Code: Which Coding Agent to Use](https://openbot.run/compare/codex-vs-claude-code) 4. ### 4.Cursor Best for: A full code editor with agents, and models from many providers. Cursor is a code editor built from VS Code, with an Agents Window, a command line and cloud agents. SpaceX bought it in August 2026, and Grok 4.7 is now its flagship model. Bugbot reviews pull requests, and an iPhone app follows cloud agents. - [Cursor vs Claude Code: Editor or Terminal Agent](https://openbot.run/compare/cursor-vs-claude-code) 5. ### 5.Antigravity Best for: Gemini models, with a free plan to start. Google's agent platform has a desktop app, a command line and an IDE for macOS, Windows and Linux, plus extensions for other editors. Its command line replaced Gemini CLI for personal plans. A browser agent drives Chrome for you. - [Claude Code vs Antigravity: Claude or Gemini Agent](https://openbot.run/compare/claude-code-vs-antigravity) 6. ### 6.Devin Best for: An autonomous software engineer in the cloud. Cognition's agent works in cloud VMs that keep going while your laptop is closed. You start it from Slack, Microsoft Teams, GitHub, Linear or Jira, and Devin Desktop and the Devin CLI run it on your computer. It doesn't take your own API keys. - [OpenBot vs Devin: A Local Devin Alternative](https://openbot.run/compare/devin) 7. ### 7.OpenClaw Best for: An open-source personal assistant in your messaging apps. OpenClaw is an open-source assistant that talks to you in WhatsApp, Telegram, Slack, Signal, iMessage or Discord. It runs on a computer you choose, with apps for desktop, iPhone and Android, and a big community writes plugins and skills for it. - [OpenBot vs OpenClaw: An OpenClaw Alternative](https://openbot.run/compare/openclaw) 8. ### 8.Hermes Agent Best for: An open-source agent that learns skills over time. Nous Research's agent writes its own skills and searches its past conversations. You talk to it in Telegram, WhatsApp, Signal, Slack or email, or in its desktop app, terminal UI or web dashboard. - [OpenBot vs Hermes Agent: A Hermes Agent Alternative](https://openbot.run/compare/hermes-agent) 9. ### 9.Claude Cowork Best for: Claude as an agent for knowledge work, in Anthropic's cloud. Anthropic's agent keeps working in the cloud while your computer is off, and it runs scheduled tasks. You start and follow tasks in the Claude desktop app, on the web, or in the Claude apps for iPhone and Android. - [OpenBot vs Claude Cowork: A Local Alternative](https://openbot.run/compare/claude-cowork) 10. ### 10.Manus Best for: Research, websites and slides from one prompt. Manus is a general agent that works on its own cloud computer and runs many agents in parallel. You hand it tasks in its apps, on the web, or through Telegram, WhatsApp Business, LINE, Slack or email. It doesn't take your own AI plan or API key. - [OpenBot vs Manus: A Local Manus Alternative](https://openbot.run/compare/manus) 11. ### 11.ChatGPT dots Best for: One always-on agent for ChatGPT Pro and Business Premium users. A dot is an OpenAI agent with its own cloud computer running Linux and Chrome. You set it up in the ChatGPT desktop app or on the web, and talk to it in the ChatGPT mobile app, Slack or Microsoft Teams. On Pro, it isn't available in the European Economic Area, Switzerland or the UK. - [OpenBot vs ChatGPT dots: A Local Alternative](https://openbot.run/compare/chatgpt-dots) 12. ### 12.Grok Bot Best for: Bots on a cloud computer, for Cursor and SuperGrok users. xAI's agent app runs your Bots on a cloud computer, so your own machine can stay off. If you approve it, it can also run tasks on your computer. There are apps for macOS, Windows, Linux, iPhone, iPad and Android. - [OpenBot vs Grok Bot: A Local Grok Bot Alternative](https://openbot.run/compare/grok-bot) 13. ### 13.Muse Best for: A personal assistant for email and purchases. Meta's personal agent keeps working after you close the app. You reach it in its iPhone, Android and Mac apps, on the web, or in WhatsApp. It's only for adults (18+) in the US and Canada. OpenBot runs Muse Code, Meta's coding agent, not this app. - [OpenBot vs Muse: A Local Alternative to Meta's Muse](https://openbot.run/compare/muse) ## Questions ### What is the best AI agent app? Depends where you want the agent to work and which AI plan you pay for. For coding with one model family, it's Claude Code, Codex, Cursor or Antigravity. On the Artificial Analysis index, Claude Code scores highest and Codex costs least per task. For an agent that works while your computer is off, Devin, Manus, Claude Cowork and ChatGPT dots run in the cloud. OpenBot runs several coding agents as one team on your own computer. ### Which AI agent apps are free? OpenBot is free for noncommercial use, and OpenClaw and Hermes Agent are free under the MIT License. With all three, you still pay your model provider. Cursor, Antigravity, Devin, Manus and Muse have free plans. Codex comes with ChatGPT Free and Go, but only with GPT-6 Luna in the desktop app and no cloud tasks. Claude Code, Claude Cowork and ChatGPT dots need a paid plan. ### Which AI agents run on my own computer? OpenBot, OpenClaw and Hermes Agent run on a computer you choose. Claude Code, Codex, Cursor and Antigravity run on your computer, and most also have a cloud option. ChatGPT dots, Manus, Muse, Grok Bot and Devin can use your computer too, but they work on a cloud computer by default. ### Can I use more than one AI agent at a time? Yes. In OpenBot, Codex, Claude Code, Antigravity, Grok CLI, Cursor CLI, Cline, OpenCode, Pi and Muse Code work as agents on one team. Each one uses your own plan for its provider, and they hand work to each other in shared channels. ### Why is OpenBot first on this list? Because we make it. Every other entry links to a comparison with its sources and the date we checked them. ## Sources Every statement about these apps comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Claude Code overview](https://code.claude.com/docs/en/overview) - [Claude pricing](https://claude.com/pricing) - [Codex documentation](https://learn.chatgpt.com/docs) - [Codex pricing](https://learn.chatgpt.com/docs/pricing) - [Cursor pricing](https://cursor.com/pricing) - [Cursor joins SpaceX](https://cursor.com/blog/joining-spacex) - [Antigravity](https://antigravity.google/) - [Antigravity pricing](https://antigravity.google/pricing) - [Devin pricing](https://devin.ai/pricing) - [OpenClaw on GitHub](https://github.com/openclaw/openclaw) - [Hermes Agent on GitHub](https://github.com/NousResearch/hermes-agent) - [Claude Cowork](https://claude.com/product/cowork) - [Manus pricing](https://manus.im/pricing) - [Dots in ChatGPT](https://chatgpt.com/features/dots/) - [Grok Bot overview](https://docs.x.ai/grok-bot/overview) - [Meta: Introducing Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) - [Muse subscription plans](https://www.meta.com/help/subscriptions/1021145227643680/) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) - [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents) - [Artificial Analysis coding agent methodology](https://artificialanalysis.ai/methodology/coding-agents-benchmarking) --- Source: https://openbot.run/compare/codex-vs-claude-code [All comparisons](https://openbot.run/compare) Comparison # Codex vs Claude Code: Which Coding Agent to Use Codex vs Claude Code: benchmark scores, cost per task, plans, sandboxes and privacy, from official sources. Or run both as one team in OpenBot. Checked Oct 3, 2026 Sources ## The short answer Pick Codex if you pay for ChatGPT and care what each task costs. Pick Claude Code if you pay for Claude and want the top benchmark score, and can wait for it. Both start at $20 a month, and both run on your computer or in the cloud. ### Choose Codex if - You already pay for ChatGPT. - Cost per task matters: GPT-6.1 Sol scores 63 on the Artificial Analysis index for about $1 a task. - The sandbox should be on from the first run, with no network. - You'd like to read the source. The command line is Apache-2.0. ### Choose Claude Code if - You already pay for Claude. - You want the best score on the index: 68, with Sonnet 5.5 at max effort. - Your company buys Claude through Amazon Bedrock, Google Cloud or Microsoft Foundry. ## At a glance Codex wins 2 of 9 topics, and Claude Code wins 1. The rest are a draw. Codex and Claude Code compared, as checked on Oct 3, 2026 | Topic | Codex | Claude Code | | --- | --- | --- | | Models | OpenAI models: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna. | Claude models: Opus 5.5 is the default, with Sonnet, Haiku and Fable models. | | Where agents work | On your computer, or in cloud tasks that run in virtual machines that OpenAI manages. | On your computer, or in cloud sessions that run in virtual machines that Anthropic manages. | | Sandbox | Better: An operating-system sandbox is on by default: no network, and writes only in the workspace. | The operating-system sandbox is off by default. Turn it on with /sandbox. Not on native Windows. | | Plans and price | Plus $20 a month, Pro $100, $200 or $500, and Business from $20 a user. Free and Go get GPT-6 Luna in the desktop app, as it rolls out. An API key also works, with no cloud features. | Pro $20 a month, or $17 billed yearly. Max $100 or $200, and Team seats $25, or $20 billed yearly. The free plan does not include Claude Code. An API key also works. | | Parallel work | Subagents, Git worktrees for parallel chats, and many cloud tasks at the same time. | Subagents, background sessions, Git worktrees and many cloud sessions. Agent teams are experimental and off by default. | | From your phone | The ChatGPT app for iPhone controls tasks on your Mac or Windows PC. | The Claude app for iPhone and Android follows cloud sessions and controls sessions on your computer. | | Your data | On Plus and Pro, OpenAI can use your chats to improve its models unless you turn training off. Business, Enterprise and the API: no training by default. | Better: On Pro and Max, Anthropic trains on your chats only when you turn the setting on. Team, Enterprise and the API: no training by default. | | Apps | A command line for macOS, Linux and Windows. Codex in the ChatGPT desktop app for macOS and Windows, with Linux in preview. Extensions for VS Code, Cursor and Windsurf. | A command line for macOS, Linux and Windows. A desktop app for macOS and Windows, with Linux in beta. Extensions for VS Code and JetBrains IDEs, and the web. | | Source code | Better: The command line is open source under Apache-2.0. The IDE extension and the cloud are not. | Proprietary. The GitHub repository holds plugins, examples and issues, not the source. | ## By the numbers Claude Code with Sonnet 5.5 at max effort has the top score, 68. It's also the slowest and most expensive run here: $14.19 and 87 minutes a task. Codex with GPT-6.1 Sol at xhigh scores 63, the same as Sonnet 5.5 at xhigh, for $1.04 a task instead of $3.33, and in a bit more than half the time. Artificial Analysis Coding Agent Index, read on Oct 3, 2026 | Agent and model | Index score Higher is better | API cost per task Lower is better | Time per task Lower is better | | --- | --- | --- | --- | | Claude Code Sonnet 5.5 (max) | 68 | $14.19 | 87 min | | Claude Code Opus 5.5 (max) | 66 | $13.04 | 65 min | | Claude Code Sonnet 5.5 (xhigh) | 63 | $3.33 | 27 min | | Codex GPT-6.1 Sol (xhigh) | 63 | $1.04 | 16 min | | Codex GPT-6 Astra (max) | 62 | $7.47 | 29 min | | Codex GPT-6.1 Sol (medium) | 61 | $0.70 | 11 min | Data from the [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents), read on Oct 3, 2026. Cost is what a task costs at pay-per-token API prices, not a plan price. Each row is one agent with one model, tested the [way Artificial Analysis describes](https://artificialanalysis.ai/methodology/coding-agents-benchmarking). ## You don't have to pick one OpenBot runs both on your computer, each signed in with your own plan: Codex with ChatGPT, Claude Code with Claude. Give each one a job. They pass work to each other in a shared channel, and you can follow along from your phone. Move an agent from one to the other and it keeps its role, workspace and conversation. When one plan hits its limit, the agent can move to a backup model on the other. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. - [Set up Codex in OpenBot](https://openbot.run/providers/codex) - [Set up Claude Code in OpenBot](https://openbot.run/providers/claude-code) Codex is OpenAI's coding agent. Claude Code is Anthropic's. They do the same job: read your code, edit files and run commands, on your computer or in their maker's cloud. Both read rules from your repo, use MCP servers, skills and hooks, and review pull requests on GitHub. The real difference is whose models you get, and which subscription you already have. ## Models and the plans you pay for ### Codex Codex is built for OpenAI's models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. OpenAI recommends GPT-6.1 Sol for hard coding work. Plus is $20 a month, Pro is $100, $200 or $500, and Business is $20 a seat billed yearly. Free and Go get GPT-6 Luna in the desktop app as it rolls out, but no cloud tasks. Limits reset every five hours, and weekly caps can apply. An API key works too, at API prices, but you lose cloud features like GitHub review. ### Claude Code Claude Code is built for Claude's models, with Opus 5.5 as the default on paid plans. Pro is $20 a month ($17 billed yearly), Max is $100 or $200, and Team seats are $25 ($20 yearly). The free Claude plan doesn't include Claude Code. Limits reset every five hours, and paid plans have a weekly cap that Claude and Claude Code share. It also works with an Anthropic API key, Amazon Bedrock, Google Cloud and Microsoft Foundry. ## Where the work happens ### Codex Better Codex sandboxes local work from the first run: Seatbelt on macOS, bubblewrap on Linux and a native sandbox on Windows. The network is off and writes stay in the workspace. Cloud tasks run in VMs that OpenAI manages, and they keep going while your laptop sleeps. Of the two, this is the safer default. ### Claude Code Claude Code runs on your computer by default. Its sandbox (Seatbelt on macOS, bubblewrap on Linux) stays off until you turn it on, and native Windows doesn't have one. Cloud sessions run in an isolated VM that Anthropic manages, or in your own environment. Your GitHub credentials never go into that VM. ## Parallel work and teams ### Codex Codex runs subagents in parallel, and its Ultra reasoning mode leans on them. Git worktrees let you run several chats in one project, and each cloud task gets its own workspace. You can start work from GitHub, Slack and Linear. GitLab is in beta. ### Claude Code Claude Code has subagents with their own context, and background sessions through `claude agents`. Worktrees keep parallel work apart, and each cloud session runs on its own. Agent teams, where several sessions work together, are still experimental and off by default. You can start work from GitHub with @claude, or from Slack. ## Data and privacy ### Codex On Plus and Pro, OpenAI can train on your chats unless you turn that off in ChatGPT's data controls, which cover Codex too. Business, Enterprise, Edu and the API aren't used for training by default. Enterprise and Edu also get retention and data-residency controls. ### Claude Code Better On Free, Pro and Max, Anthropic trains on your data only if you turn the setting on. With it on, data is kept for 5 years. With it off, 30 days. Team, Enterprise and the API aren't used for training unless you opt in, and Enterprise can get zero data retention. Opt-in beats opt-out, so Claude Code takes this one. ## Questions ### Which is better for coding, Codex or Claude Code? On the Artificial Analysis Coding Agent Index, Claude Code with Sonnet 5.5 at max effort scores highest, 68. Codex with GPT-6.1 Sol at xhigh scores 63 for $1.04 a task, against $14.19 for that Claude run. Past the numbers, the plan you already pay for usually decides it. ### Can I use Codex and Claude Code together? Yes. In OpenBot, each one is an agent with its own job, signed in with your own ChatGPT or Claude plan. They hand work to each other in a shared channel on your computer, and you follow them from the OpenBot app on your phone. ### How much do Codex and Claude Code cost? Both start at $20 a month: ChatGPT Plus for Codex, Claude Pro for Claude Code. Claude Pro drops to $17 a month if you pay yearly. Higher limits cost $100 or $200 a month on both, and ChatGPT Pro also has a $500 tier. ChatGPT Free and Go get GPT-6 Luna in the desktop app as it rolls out. The free Claude plan doesn't include Claude Code. ### Is Codex open source? Is Claude Code? The Codex command line is, under Apache-2.0, at openai/codex on GitHub. The IDE extension and Codex Cloud aren't. Claude Code is proprietary. Its GitHub repo holds plugins, examples and issues, not the source. ### Do Codex and Claude Code train on my code? On personal plans, it comes down to one setting. ChatGPT Plus and Pro train on your chats unless you turn it off. Claude Pro and Max train only if you turn it on. On business plans and the API, neither trains on your data by default. ### Can I use an API key instead of a plan? Yes, with both. Codex takes an OpenAI API key in the command line, SDK and IDE extension, at API prices, but cloud tasks need a ChatGPT sign-in. Claude Code takes an Anthropic API key, and also runs through Amazon Bedrock, Google Cloud and Microsoft Foundry. ## Sources Every statement about Codex and Claude Code comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Codex documentation](https://learn.chatgpt.com/docs) - [Codex pricing](https://learn.chatgpt.com/docs/pricing) - [Codex models](https://learn.chatgpt.com/docs/models) - [Codex sandboxing](https://learn.chatgpt.com/docs/sandboxing) - [Codex cloud environments](https://learn.chatgpt.com/docs/environments/cloud-environments) - [Codex subagents](https://learn.chatgpt.com/docs/agent-configuration/subagents) - [Codex Git worktrees](https://learn.chatgpt.com/docs/environments/git-worktrees) - [Codex remote](https://learn.chatgpt.com/docs/remote) - [Codex IDE extension](https://learn.chatgpt.com/docs/codex/ide) - [Codex open source](https://learn.chatgpt.com/docs/open-source) - [Codex authentication](https://learn.chatgpt.com/docs/auth) - [What's new in Codex](https://learn.chatgpt.com/docs/whats-new) - [Using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan) - [openai/codex on GitHub](https://github.com/openai/codex) - [Claude Code overview](https://code.claude.com/docs/en/overview) - [Claude Code setup](https://code.claude.com/docs/en/setup) - [Claude Code model configuration](https://code.claude.com/docs/en/model-config) - [Claude Code sandboxing](https://code.claude.com/docs/en/sandboxing) - [Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web) - [Claude Code agent teams](https://code.claude.com/docs/en/agent-teams) - [Claude Code worktrees](https://code.claude.com/docs/en/worktrees) - [Claude Code on mobile](https://code.claude.com/docs/en/mobile) - [Claude Code desktop](https://code.claude.com/docs/en/desktop) - [Claude Code data usage](https://code.claude.com/docs/en/data-usage) - [Claude pricing](https://claude.com/pricing) - [Use Claude Code with your Team or Enterprise plan](https://support.claude.com/en/articles/11845131-use-claude-code-with-your-team-or-enterprise-plan) - [Claude Code license](https://github.com/anthropics/claude-code/blob/main/LICENSE.md) - [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents) - [Artificial Analysis coding agent methodology](https://artificialanalysis.ai/methodology/coding-agents-benchmarking) --- Source: https://openbot.run/compare/claude-code-vs-antigravity [All comparisons](https://openbot.run/compare) Comparison # Claude Code vs Antigravity: Claude or Gemini Agent Claude Code vs Google Antigravity: benchmark scores, cost per task, plans, apps and privacy, from official sources. Or run both in OpenBot. Checked Oct 3, 2026 Sources ## The short answer Pick Claude Code if you pay for Claude and want the stronger agent today. Pick Antigravity if you want Gemini, a free plan to start, and Claude models thrown in on Google AI Pro or Ultra. Antigravity's best score comes from a Gemini model that isn't public yet. ### Choose Claude Code if - You already pay for Claude. - You want the top score on the Artificial Analysis index: 68, with Sonnet 5.5 at max effort. - Some work should run in Anthropic's cloud while your computer is off. - You'd like a real phone app, on iPhone and Android, to follow sessions. ### Choose Antigravity if - You pay for Google AI Pro or Ultra, or want to start free. - Gemini is your default, but you'd still like Claude Sonnet 5.5 and Opus 5.5 on Pro and Ultra. - You'd like a desktop app built for agents, with a browser agent that drives Chrome. ## At a glance Claude Code wins 3 of 10 topics, and Antigravity wins 3. The rest are a draw. Claude Code and Antigravity compared, as checked on Oct 3, 2026 | Topic | Claude Code | Antigravity | | --- | --- | --- | | Models | Claude models: Opus 5.5 is the default, with Sonnet, Haiku and Fable models. | Better: Gemini 3.8 Flash is the default, with other Gemini models. Claude Sonnet 5.5 and Opus 5.5 on Pro and Ultra. | | Where agents work | On your computer, or in cloud sessions that run in virtual machines that Anthropic manages. | On your computer. Remote Control connects a browser to it. A cloud agent is a separate Gemini API product. | | Sandbox | The operating-system sandbox is off by default. Turn it on with /sandbox. Not on native Windows. | Better: The terminal sandbox is on by default on macOS and Linux. Windows does not have it yet. | | Plans and price | Pro $20 a month, or $17 billed yearly. Max $100 or $200, and Team seats $25, or $20 billed yearly. The free plan does not include Claude Code. | Better: A free plan with weekly limits. Google AI Pro $19.99 a month, and AI Ultra $99.99 or $199.99. Enterprise through Google Cloud. | | Own API key | Better: An Anthropic API key works, and so do Amazon Bedrock, Google Cloud and Microsoft Foundry. | No own key or endpoint for more limits. The IDE extensions can sign in with a Gemini API key. | | Parallel work | Subagents, background sessions, Git worktrees and many cloud sessions. Agent teams are experimental and off by default. | Parallel local subagents, several workspaces and Git worktrees, and scheduled tasks. | | From your phone | Better: The Claude app for iPhone and Android follows cloud sessions and controls sessions on your computer. | Remote Control in any web browser. You can add it to your phone's home screen. No native app is stated. | | Your data | Better: On Pro and Max, Anthropic trains on your chats only when you turn the setting on. Team, Enterprise and the API: no training by default. | On personal plans, Google can use your interactions to improve its products, and people can review them. You can opt out. Enterprise: no training. | | Apps | A command line for macOS, Linux and Windows. A desktop app for macOS and Windows, with Linux in beta. Extensions for VS Code and JetBrains IDEs, and the web. | The Antigravity 2.0 desktop app and command line for macOS, Windows and Linux, and the Antigravity IDE. Extensions for VS Code, Visual Studio, JetBrains IDEs, Zed and Xcode. | | Source code | Proprietary. | Proprietary. Gemini CLI, which it replaces for personal plans, stays open source. | ## By the numbers Claude Code leads with 68. Antigravity's best run scores 64, but it uses Gemini 4 Argon, which you can't use yet. Gemini 3.8 Flash, Antigravity's default model, scores 42 for $2.47 and 12 minutes a task. Artificial Analysis tested it through the Antigravity SDK, not the CLI, so read that row with care. Artificial Analysis Coding Agent Index, read on Oct 3, 2026 | Agent and model | Index score Higher is better | API cost per task Lower is better | Time per task Lower is better | | --- | --- | --- | --- | | Claude Code Sonnet 5.5 (max) | 68 | $14.19 | 87 min | | Claude Code Opus 5.5 (max) | 66 | $13.04 | 65 min | | Claude Code Sonnet 5.5 (xhigh) | 63 | $3.33 | 27 min | | Antigravity CLI Gemini 4 Argon (high)Not public yet | 64 | $5.84 | 35 min | | Antigravity SDK Gemini 3.8 Flash (high)Tested with the SDK, not the CLI | 42 | $2.47 | 12 min | Data from the [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents), read on Oct 3, 2026. Cost is what a task costs at pay-per-token API prices, not a plan price. Each row is one agent with one model, tested the [way Artificial Analysis describes](https://artificialanalysis.ai/methodology/coding-agents-benchmarking). ## You don't have to pick one OpenBot runs Claude Code and Gemini on your computer, each signed in with your own plan: Claude Code with Claude, Gemini with Google AI Pro or Ultra through Google's Antigravity CLI. Give each one a job. They pass work to each other in a shared channel, and you can follow along from your phone. When one plan hits its limit, the agent can move to a backup model on the other. Move an agent from one to the other and it keeps its role, workspace and conversation. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - ChatGPT Your ChatGPT plan, through Codex. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. - [Set up Claude Code in OpenBot](https://openbot.run/providers/claude-code) - [Set up Gemini in OpenBot](https://openbot.run/providers/gemini) Claude Code is Anthropic's coding agent. Antigravity is Google's agent platform. Both read your code, edit files and run commands on your computer, work in parallel, and read rules, skills, hooks and MCP servers from your project. Why Antigravity and not Gemini CLI? On 18 June 2026, Gemini CLI stopped serving Google AI Pro, Ultra and free users, and Google moved them to Antigravity CLI. On a personal Google plan, Antigravity is the agent you'd actually use. ## Models and the plans you pay for ### Claude Code Claude Code is built for Claude's models, with Opus 5.5 as the default on paid plans. Pro is $20 a month ($17 billed yearly), Max is $100 or $200, and Team seats are $25 ($20 yearly). The free Claude plan doesn't include Claude Code. It also works with an Anthropic API key, Amazon Bedrock, Google Cloud and Microsoft Foundry. ### Antigravity Better Antigravity uses Gemini 3.8 Flash for local agents by default. Gemini 3.7 Flash, 3.6 Flash and 3.1 Pro are on every plan, and Claude Sonnet 5.5 and Opus 5.5 come with paid Pro and Ultra. The free plan has weekly limits. Google AI Pro is $19.99 a month and AI Ultra is $99.99 or $199.99 in the US. Limits reset every five hours, up to a weekly cap, and your own API key can't raise them. Two model families for one price is hard to beat. ## Where the work happens ### Claude Code Claude Code runs on your computer by default. Its sandbox stays off until you turn it on, and native Windows doesn't have one. Cloud sessions run in an isolated VM that Anthropic manages. From the Claude phone app, you follow cloud sessions and control the ones on your computer. ### Antigravity Antigravity runs its agents on your computer, in a terminal sandbox that's on by default on macOS and Linux. Remote Control lets you drive a session from any web browser, but the agent still runs on your machine. Google's only cloud version is Managed Agents, a separate Gemini API product. ## Parallel work and teams ### Claude Code Claude Code has subagents with their own context, and background sessions through `claude agents`. Worktrees keep parallel work apart, and each cloud session runs on its own. Agent teams are still experimental and off by default. You can start work from GitHub with @claude, or from Slack. ### Antigravity Antigravity 2.0 works across several workspaces and Git worktrees, and splits a task between parallel local subagents. It can run tasks on a schedule. In the command line, /agents shows the subagents working in the background. ## Data and privacy ### Claude Code Better On Free, Pro and Max, Anthropic trains on your data only if you turn the setting on. With it on, data is kept for 5 years. With it off, 30 days. Team, Enterprise and the API aren't used for training unless you opt in. ### Antigravity On personal plans, Google records your interactions, uses them to improve its products and machine learning, and lets human reviewers read them. You can opt out in settings. On Enterprise, Google never trains its foundation models on your code, prompts or transcripts, and you can keep data in the US or the EU. Google gives no retention period. Opt-in beats opt-out, so Claude Code takes this one. ## Questions ### Is Antigravity the same as Gemini CLI? No. Gemini CLI is Google's open-source command line. Antigravity is Google's agent platform: a desktop app, the Antigravity CLI, an IDE and IDE extensions. On 18 June 2026, Gemini CLI stopped serving Google AI Pro, Ultra and free users, and Google pointed them to Antigravity CLI. Paid API keys and Gemini Code Assist licences still work with Gemini CLI. ### Which is better for coding, Claude Code or Antigravity? On the Artificial Analysis Coding Agent Index, Claude Code scores 68 with Sonnet 5.5 at max effort. Antigravity reaches 64 only with Gemini 4 Argon, which isn't public yet. Antigravity is the better deal if you want a free plan, or Gemini and Claude on one subscription. ### Can I use Claude Code and Antigravity together? Yes. In OpenBot, Claude Code and Gemini are agents on one team, each signed in with your own Claude or Google AI plan. OpenBot runs Gemini through Google's Antigravity CLI. They hand work to each other in a shared channel on your computer. ### Can I use Claude models in Antigravity? Yes, on paid plans. Claude Sonnet 5.5 and Opus 5.5 come with Google AI Pro (not the trial) and Ultra. They're not on the free plan or Enterprise. ### How much do Claude Code and Antigravity cost? Claude Code needs a paid Claude plan: Pro is $20 a month ($17 billed yearly), and Max is $100 or $200. Antigravity has a free plan with weekly limits. Google AI Pro is $19.99 a month in the US, and AI Ultra is $99.99 or $199.99. ### Do Claude Code and Antigravity train on my code? On personal plans, Anthropic trains on your data only if you turn it on. Google can use your Antigravity interactions to improve its products unless you opt out. On business plans, neither trains on your data by default. ## Sources Every statement about Claude Code and Antigravity comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Claude Code overview](https://code.claude.com/docs/en/overview) - [Claude Code setup](https://code.claude.com/docs/en/setup) - [Claude Code model configuration](https://code.claude.com/docs/en/model-config) - [Claude Code sandboxing](https://code.claude.com/docs/en/sandboxing) - [Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web) - [Claude Code agent teams](https://code.claude.com/docs/en/agent-teams) - [Claude Code on mobile](https://code.claude.com/docs/en/mobile) - [Claude Code desktop](https://code.claude.com/docs/en/desktop) - [Claude Code data usage](https://code.claude.com/docs/en/data-usage) - [Claude pricing](https://claude.com/pricing) - [Claude Code license](https://github.com/anthropics/claude-code/blob/main/LICENSE.md) - [Antigravity](https://antigravity.google/) - [Antigravity documentation](https://antigravity.google/docs/home/) - [Antigravity overview](https://antigravity.google/docs/overview/) - [Antigravity models](https://antigravity.google/docs/models/) - [Antigravity plans](https://antigravity.google/docs/plans/) - [Antigravity pricing](https://antigravity.google/pricing) - [Antigravity sandbox](https://antigravity.google/docs/sandbox/) - [Antigravity Remote Control](https://antigravity.google/docs/remote-control/) - [Antigravity CLI reference](https://antigravity.google/docs/cli/reference/) - [Antigravity enterprise](https://antigravity.google/docs/enterprise/) - [Antigravity FAQ](https://antigravity.google/docs/faq/) - [Antigravity terms](https://antigravity.google/terms) - [Antigravity download](https://antigravity.google/download) - [Google AI subscriptions](https://gemini.google/subscriptions/?hl=en) - [Transitioning Gemini CLI to Antigravity CLI](https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/) - [Antigravity in the ACP registry](https://github.com/agentclientprotocol/registry/tree/main/antigravity-acp) - [Gemini API managed agents](https://ai.google.dev/gemini-api/docs/antigravity-agent) - [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents) - [Artificial Analysis coding agent methodology](https://artificialanalysis.ai/methodology/coding-agents-benchmarking) --- Source: https://openbot.run/compare/cursor-vs-claude-code [All comparisons](https://openbot.run/compare) Comparison # Cursor vs Claude Code: Editor or Terminal Agent Cursor vs Claude Code: an editor with many models, or Claude's agent in your terminal. Plans, cloud agents and privacy, from official sources. Checked Oct 3, 2026 Sources ## The short answer Pick Cursor if you want an editor with agents built in and models from several labs on one bill. Pick Claude Code if you want Claude's agent in the terminal and editor you already use. Both start at $20 a month, and both run agents on your computer or in the cloud. ### Choose Cursor if - You want a full editor, built from VS Code, with agents, completions and review in one place. - Grok, Claude, GPT, Gemini and Cursor's own models on one plan sounds good. - Bugbot should review your pull requests on GitHub, GitLab, Bitbucket or Azure DevOps. ### Choose Claude Code if - You already pay for Claude. - You're happy with your terminal and editor, VS Code or JetBrains, and don't want to switch. - Your company buys Claude through Amazon Bedrock, Google Cloud or Microsoft Foundry. ## At a glance Cursor wins 1 of 9 topics, and Claude Code wins 1. The rest are a draw. Cursor and Claude Code compared, as checked on Oct 3, 2026 | Topic | Cursor | Claude Code | | --- | --- | --- | | Models | Better: Cursor's own models, such as Grok 4.7 and Composer 2.5, and models from Anthropic, OpenAI, Google and Meta. | Claude models: Opus 5.5 is the default, with Sonnet, Haiku and Fable models. | | Where agents work | On your computer, in the editor or the command line, or in cloud virtual machines. Cloud agents can also run on machines that you host. | On your computer, or in cloud sessions that run in virtual machines that Anthropic manages, or in your own environment. | | Plans and price | Hobby is free. Pro $20 a month, Pro+ $60 and Ultra $200. Teams $40 or $120 a user. Use beyond the plan is billed at API prices. | Pro $20 a month, or $17 billed yearly. Max $100 or $200, and Team seats $25, or $20 billed yearly. The free plan does not include Claude Code. | | Own API key | Keys from OpenAI, Anthropic, Google, Azure OpenAI and Amazon Bedrock, for chat models only. The command line also takes a Cursor API key. | An Anthropic API key works, and so do Amazon Bedrock, Google Cloud and Microsoft Foundry. | | Parallel work | Parallel agents in the Agents Window, Git worktrees, and as many cloud agents as you want. Projects, in beta, lets one agent coordinate others. | Subagents, background sessions, Git worktrees and many cloud sessions. Agent teams are experimental and off by default. | | From your phone | Cursor for iPhone and iPad, on iOS 26 or later, for cloud agents. Android is planned; until then, a web app. | Better: The Claude app for iPhone and Android follows cloud sessions and controls sessions on your computer. | | Your data | With Privacy Mode on, Cursor does not train on your data. With it off, Cursor can use your code and prompts to train. Cursor does not state the default for personal plans. | On Pro and Max, Anthropic trains on your chats only when you turn the setting on. Team, Enterprise and the API: no training by default. | | Apps | The Cursor editor for macOS, Windows and Linux, a command line, the web and an iPhone app. Integrations with JetBrains IDEs and Xcode. | A command line for macOS, Linux and Windows. A desktop app for macOS and Windows, with Linux in beta. Extensions for VS Code, its forks and JetBrains IDEs. | | Source code | Proprietary. The editor is a fork of VS Code. | Proprietary. | ## You don't have to pick one OpenBot runs Cursor CLI and Claude Code on your computer, each signed in with your own plan: Cursor CLI with your Cursor plan or a Cursor API key, Claude Code with Claude. Give each one a job. They pass work to each other in a shared channel, and you can follow along from your phone. When one plan hits its limit, the agent can move to a backup model on the other. Move an agent from one to the other and it keeps its role, workspace and conversation. - Claude Your Claude plan, through Claude Code. - Cursor Your Cursor plan, or a Cursor API key. - ChatGPT Your ChatGPT plan, through Codex. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. - [Set up Cursor in OpenBot](https://openbot.run/providers/cursor) - [Set up Claude Code in OpenBot](https://openbot.run/providers/claude-code) Cursor is an AI code editor from Anysphere, which SpaceX bought in August 2026. Claude Code is Anthropic's coding agent for the terminal, the IDE, the desktop and the web. Both run agents that edit files and run commands, locally or in the cloud, and both read AGENTS.md-style rules, MCP servers, hooks and skills. The difference is the shape of the tool. Cursor is the editor, with many models. Claude Code plugs into the tools you already have, with Claude's models. ## Models and the plans you pay for ### Cursor Better Cursor's own pool has Grok 4.7, 4.6 and 4.5 and Composer 2.5, and Cursor calls Grok 4.7 its flagship. Claude, GPT, Gemini and Muse models are there too, at API prices. Hobby is free, Pro is $20 a month, Pro+ $60 and Ultra $200, and Teams is $40 or $120 a user. Each plan includes some usage, and anything past it is billed at API prices. Your own keys from OpenAI, Anthropic, Google, Azure OpenAI and Amazon Bedrock work for chat models. If you want to switch models freely, Cursor wins. ### Claude Code Claude Code is built for Claude's models, with Opus 5.5 as the default on paid plans. Pro is $20 a month ($17 billed yearly), Max is $100 or $200, and Team seats are $25 ($20 yearly). The free Claude plan doesn't include Claude Code. Limits reset every five hours, and paid plans have weekly caps too. It also works with an Anthropic API key, Amazon Bedrock, Google Cloud and Microsoft Foundry. ## Editor or agent ### Cursor Cursor is a code editor, a fork of VS Code, with extensions from Open VSX. The Agents Window runs agents next to the editor: locally, in worktrees, over SSH or in the cloud. The Cursor CLI, `agent`, runs the same agent in a terminal and can hand a task off to a cloud agent. ### Claude Code Claude Code started in the terminal, and it doesn't replace your editor. Its extension works in VS Code and its forks, Cursor included, and its plugin works in JetBrains IDEs. The desktop app runs parallel sessions, locally, in the cloud or over SSH, and the web app lives at claude.ai/code. ## Parallel work and teams ### Cursor The Agents Window runs agents in parallel, each in its own Git worktree if you like. Cloud agents run in isolated VMs, as many as you want at once, and you can start them from Slack, Microsoft Teams, Linear, Jira or a GitHub comment. Projects, in beta since September 2026, lets one coordinator agent hand work to others. ### Claude Code Claude Code has subagents with their own context, and background sessions through `claude agents`. Worktrees keep parallel work apart, and each cloud session runs on its own. Agent teams, where several sessions work together, are still experimental and off by default. You can start work from GitHub with @claude, or from Slack. ## Data and privacy ### Cursor With Privacy Mode on, Cursor doesn't train on your data and has zero data retention deals with its model providers. With it off, Cursor can store your code and prompts and train on them. Privacy Mode is on by default for Enterprise teams. For personal plans, Cursor doesn't say. Requests go through Cursor's servers, even with your own API key. Cloud agents are the only feature that stores your code, and Cursor deletes it when the agent finishes. ### Claude Code On Free, Pro and Max, Anthropic trains on your data only if you turn the setting on. With it on, data is kept for 5 years. With it off, 30 days. Team, Enterprise and the API aren't used for training unless you opt in, and standard retention is 30 days. Here you at least know the default. ## Questions ### Which is better for coding, Cursor or Claude Code? Depends how you work. Cursor is a full editor with agents in it and models from many labs. Claude Code is an agent that lives in your terminal and plugs into the editor you already use, with Claude's models. Both run agents locally and in the cloud. ### Is there a benchmark for Cursor and Claude Code? Not one that has both. The Artificial Analysis Coding Agent Index tests Claude Code, which scores up to 68 with Sonnet 5.5 at max effort, but it didn't include Cursor when we checked on 3 October 2026. ### Can I use Claude Code in Cursor? Yes. Claude Code's extension installs in Cursor like in any VS Code fork, and it bundles its own copy of the Claude Code command line. You can also pick Claude models inside Cursor itself, billed through your Cursor plan. ### Can I use Cursor and Claude Code together? Yes. In OpenBot, Cursor CLI and Claude Code are agents on one team, each signed in with your own Cursor or Claude plan. They hand work to each other in a shared channel on your computer, and you follow them from the OpenBot app on your phone. ### How much do Cursor and Claude Code cost? Both start at $20 a month: Cursor Pro, and Claude Pro for Claude Code. Claude Pro drops to $17 a month if you pay yearly. Cursor also has a free Hobby plan, Pro+ at $60 and Ultra at $200. Claude Max is $100 or $200. The free Claude plan doesn't include Claude Code. ### Who owns Cursor? SpaceX. Cursor's blog says SpaceX acquired it on 14 August 2026. Cursor now calls Grok 4.7 its flagship model. ### Do Cursor and Claude Code train on my code? With Cursor, it comes down to Privacy Mode: on means no training, off means Cursor can train on your code and prompts. On Claude's personal plans, Anthropic trains only if you turn the setting on. On business plans, Privacy Mode is on by default for Enterprise teams, and Anthropic doesn't train by default. ## Sources Every statement about Cursor and Claude Code comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Cursor pricing](https://cursor.com/pricing) - [Cursor models and pricing](https://cursor.com/docs/account/pricing) - [Cursor available models](https://cursor.com/help/models-and-usage/available-models) - [Cursor API keys](https://cursor.com/help/models-and-usage/api-keys) - [Cursor Agents Window](https://cursor.com/docs/agent/agents-window) - [Cursor cloud agents](https://cursor.com/docs/cloud-agents) - [Cursor on mobile](https://cursor.com/docs/cloud-agents/mobile) - [Cursor CLI](https://cursor.com/docs/cli/overview) - [Cursor Bugbot](https://cursor.com/docs/bugbot) - [Cursor downloads](https://cursor.com/downloads) - [Cursor data use](https://cursor.com/data-use) - [Cursor privacy](https://cursor.com/help/security-and-privacy/privacy) - [Cursor privacy and data governance](https://cursor.com/docs/enterprise/privacy-and-data-governance) - [Cursor 3](https://cursor.com/blog/cursor-3) - [Cursor joins SpaceX](https://cursor.com/blog/joining-spacex) - [Cursor changelog](https://cursor.com/changelog) - [Cursor terms of service](https://cursor.com/terms-of-service) - [Claude Code overview](https://code.claude.com/docs/en/overview) - [Claude Code setup](https://code.claude.com/docs/en/setup) - [Claude Code in VS Code](https://code.claude.com/docs/en/vs-code) - [Claude Code model configuration](https://code.claude.com/docs/en/model-config) - [Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web) - [Claude Code agent teams](https://code.claude.com/docs/en/agent-teams) - [Claude Code on mobile](https://code.claude.com/docs/en/mobile) - [Claude Code desktop](https://code.claude.com/docs/en/desktop) - [Claude Code data usage](https://code.claude.com/docs/en/data-usage) - [Claude pricing](https://claude.com/pricing) - [Claude Code license](https://github.com/anthropics/claude-code/blob/main/LICENSE.md) - [Artificial Analysis Coding Agent Index](https://artificialanalysis.ai/agents/coding-agents) - [Artificial Analysis coding agent methodology](https://artificialanalysis.ai/methodology/coding-agents-benchmarking) --- Source: https://openbot.run/compare/chatgpt-dots [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs ChatGPT dots: A Local Alternative OpenBot is a free, local ChatGPT dots alternative: run a team of AI agents on your own computer, in any country, with the ChatGPT plan you already pay for. Checked Oct 2, 2026 Sources ## The short answer Choose OpenBot if you want a team of agents on your own computer, in any country, with your ChatGPT plan and the other AI plans you already pay for, such as Claude, Gemini or Grok: the app is free. Choose ChatGPT dots if you pay for ChatGPT Pro or Business Premium, and you want one always-on agent in OpenAI's cloud. ### Choose OpenBot if Recommended - You live in the European Economic Area, Switzerland or the UK, where ChatGPT Pro does not include dots. - You want ChatGPT, Claude, Gemini, Grok, Cursor, Cline, Muse Code and Pi agents in one team, or your own model. - Your files and chats must stay on your own computer, not in OpenAI's cloud. - You want a free app, and source code that you can read. ### Choose ChatGPT dots if - You already pay for ChatGPT Pro or Business Premium. - You want an agent that works around the clock in the cloud, with no computer of yours to keep on. - You want to talk to your agent in Slack, Microsoft Teams or the ChatGPT apps. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **ChatGPT dots:** ChatGPT dots runs on GPT-6 Astra, an OpenAI model. One dot comes with ChatGPT Pro, from $100 a month, or with Business Premium, from $100 a seat a month. Enterprise plans get a beta. ## At a glance OpenBot is better on 8 of 14 topics, and ChatGPT dots on 0. OpenBot and ChatGPT dots compared, as checked on Oct 2, 2026 | Topic | OpenBot | ChatGPT dots | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | GPT-6 Astra, an OpenAI model. OpenAI describes no way to use other models or your own model. | | Where agents work | Better: On the computer that runs OpenBot. | On a cloud computer that OpenAI runs for each dot, with Linux and Chrome. If you allow it in the ChatGPT desktop app, a dot can also use your computer. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | A dot keeps working on its cloud computer while you are away, and runs reminders and recurring tasks on a schedule. | | Remote access | Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | After you create a dot on a computer, talk to it in the ChatGPT mobile app where mobile access is available, or in Slack and Microsoft Teams. Its chats and files are in your ChatGPT account. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Dots run only on GPT-6 Astra. OpenAI describes no other model, and no move to another model when you reach a limit. | | Logins and payments | The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | For supported sign-ins, you type your password in a secure form that the model does not see, or you take over the browser. A dot can buy with a card that you saved on a store's website, after you approve the purchase. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | You can call your dot in ChatGPT on the web, phone and computer, and talk with it. OpenAI does not say where dots process speech. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Slack and Microsoft Teams. Texting is a limited beta for Pro users in the US. OpenAI names no Discord, Telegram or WhatsApp. | | Teams | Better: Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | One dot for each person today; OpenAI plans teams of dots later. A dot works on several projects at once, can give work to other agents, and can join Slack and Teams channels. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | OpenAI keeps what a dot learns for as long as you keep the dot, and you cannot see or delete a single memory. On personal plans, the model training setting also covers the dot's work. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Needs ChatGPT Pro, from $100 a month, or Business Premium, from $100 a seat a month for 2 seats or more. The first dot is included; OpenAI gives no price for more. | | Where you can use it | Better: In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | On Pro, in the ChatGPT countries except the European Economic Area, Switzerland and the UK. On Business Premium, in all ChatGPT countries. Only for people 18 and over. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | Create a dot in the ChatGPT app for macOS or Windows, or on the web on a computer. Then talk to it in the ChatGPT mobile app, Slack or Microsoft Teams. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | A closed service that runs only in OpenAI's cloud. The auto-review rules that check a dot's actions are in the open-source Codex repository. | OpenAI started dots, its always-on agents in ChatGPT, on 29 September 2026. Each dot runs on GPT-6 Astra, has its own cloud computer, learns from your feedback, and works toward your goals around the clock, with more than 4,000 apps through plugins. OpenBot also works with your ChatGPT plan: Codex is one of its agents. The difference is where the agents run, how many work together, and where you can use them. A dot is one agent in OpenAI's cloud, and ChatGPT Pro does not include it in the European Economic Area, Switzerland or the UK. OpenBot runs a team of agents on your own computer, in any country, with ChatGPT and the other AI plans you choose. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### ChatGPT dots A dot runs on GPT-6 Astra, and OpenAI names no other model for it. Your first dot is included in ChatGPT Pro or Business Premium at no extra cost; OpenAI says that you can add more dots later, but gives no price. Chats with a dot do not count toward your ChatGPT limits, but the Codex and ChatGPT Work tasks that it starts do. For the first month after the start, dots usage does not count toward plan allowances. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### ChatGPT dots Each dot has its own cloud computer at OpenAI, with Linux and Chrome, where it browses, makes files and runs tools. It keeps working there while you are away, and it runs reminders and recurring tasks on a schedule. Access to your own computer is off at the start: you allow it, and revoke it, in the ChatGPT desktop app. A dot can then also use your local browser when its cloud browser is blocked. ## Where you can use it ### OpenBot Better OpenBot has no region lock: download it and run it in any country. The interface is in 15 languages. Hosted OpenBot servers run in the EU, and you can pay for them in euros, US dollars or Polish złoty. Each AI provider decides where its own plan works, so check that your provider is available in your country. ### ChatGPT dots On ChatGPT Pro, dots started in the ChatGPT markets except the European Economic Area, Switzerland and the UK, and OpenAI gives no date for them. On Business Premium, dots are in all the countries where ChatGPT is available. You must be 18 or older, and access can take several days to reach your account. Texting a dot is a beta for Pro users in the US only. ## Phones and remote access ### OpenBot The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### ChatGPT dots You create a dot in the ChatGPT desktop app or on the web on a computer, not on a phone. Then you talk to it in the ChatGPT mobile app, where mobile access is available, and in Slack and Microsoft Teams. Dots do not work on the mobile web. You can also talk to a dot by voice. ## How agents work as a team ### OpenBot Better In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### ChatGPT dots Today, each person starts with one primary dot, and OpenAI plans teams of dots that work together later. One dot works on several projects at the same time, and it can give work to other agents; its rules then apply to that work too. You can bring a dot into Slack and Microsoft Teams channels. Companies can preview specialist dots, each with its own identity. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### ChatGPT dots A dot keeps what it learns from chats and connected apps for as long as you keep the dot. You cannot see, change or delete a single memory: only a reset of the dot deletes them, and the files and chats that it made stay. A dot shares memory with ChatGPT. On Business, Enterprise and Edu plans, OpenAI does not train on your data by default; on personal plans, the training setting also covers what a dot does. People at OpenAI can review content even with training off. ## Questions ### Is OpenBot an alternative to ChatGPT dots? Yes, if you want your agents on your own computer. Both give you agents that keep working on your goals, and both work with your ChatGPT plan. OpenBot runs a team of agents on your computer, with Codex and the other providers you choose. A dot is one OpenAI agent on a cloud computer at OpenAI. ### Can I use ChatGPT dots in the EU or the UK? Not with ChatGPT Pro, for now. Dots started for Pro in the ChatGPT markets except the European Economic Area, Switzerland and the UK, and OpenAI gives no date for them. Business Premium includes dots in all ChatGPT countries. OpenBot has no region lock: it runs on your own computer in any country, and hosted OpenBot servers run in the EU. ### Can I use my ChatGPT plan with OpenBot? Yes. OpenBot runs Codex with your own ChatGPT plan, so the ChatGPT plan that you pay for works in OpenBot too. Next to it, Claude Code uses your Claude plan, Gemini a Google AI Pro or Ultra plan, Grok CLI your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### How much does ChatGPT dots cost? Your first dot is included in ChatGPT Pro, which costs $100, $200 or $500 a month, or in Business Premium, which costs $125 a seat a month, or $100 with annual billing, for 2 seats or more. OpenAI says that you can add more dots later, but gives no price. OpenBot is free for noncommercial use, and it uses the plans you already pay for. ### Does a dot run on my computer? No. Each dot works on its own cloud computer at OpenAI. Access to your computer is off at the start; you can allow it in the ChatGPT desktop app, and revoke it there. OpenBot agents work on your computer, or on a hosted OpenBot server that you choose. ### Do agents keep working when my laptop is closed? Yes, with both. A dot works on OpenAI's cloud computer, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### Is ChatGPT dots open source? No. Dots are a closed service that runs only in OpenAI's cloud. The auto-review rules that check a dot's actions are in the open-source Codex repository. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0. ## Sources Every statement about ChatGPT dots comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Introducing dots](https://openai.com/index/introducing-dots/) - [Safety, security and privacy in dots](https://openai.com/index/how-we-build-safety-security-and-privacy-into-dots/) - [Dots in ChatGPT](https://chatgpt.com/features/dots/) - [Getting started with your dot](https://help.openai.com/en/articles/20001530-getting-started-with-your-dot) - [Dots privacy, security and safety FAQ](https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs) - [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) - [ChatGPT Pro tiers](https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers) - [ChatGPT Business](https://help.openai.com/en/articles/8792828-chatgpt-business-overview) - [ChatGPT supported countries](https://help.openai.com/en/articles/7947663-chatgpt-supported-countries) - [DevDay 2026 recap](https://openai.com/index/devday-2026-recap/) - [Codex auto-review](https://learn.chatgpt.com/docs/sandboxing/auto-review) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/devin [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Devin: A Local Devin Alternative OpenBot is a free, local Devin alternative: run Codex, Claude Code, Gemini and Grok as a coding agent team on your computer, with plans you already have. Checked Oct 2, 2026 Sources ## The short answer Choose OpenBot if you want coding agents on your own computer that use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok: the app is free. Choose Devin if you want an autonomous software engineer that works in Cognition's cloud, with no computer of yours on, and starts from Slack, GitHub, Linear or Jira. ### Choose OpenBot if Recommended - You already pay for ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code, and want your agents to use that plan. - Your code and chats must stay on your own computer, not in Cognition's cloud. - You want no app vendor to keep your code or train models on it. - You want native apps for iPhone and Android, or a free app. ### Choose Devin if - You want agents in cloud virtual machines that keep working while your laptop is closed. - You want to start work from Slack, Microsoft Teams, GitHub, Linear or Jira. - You want many Devin sessions in parallel, each in its own virtual machine. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Devin:** Devin chooses from models by Anthropic, OpenAI, Google, Cognition and open-source labs, and you pay Devin for use: Free with a light quota, Pro $20 a month, Max $200 a month. It does not take your own API keys. ## At a glance OpenBot is better on 6 of 14 topics, and Devin on 0. OpenBot and Devin compared, as checked on Oct 2, 2026 | Topic | OpenBot | Devin | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Models from Anthropic, OpenAI, Google, Cognition and open-source labs, paid through your Devin plan. No own API keys. On paid plans, Devin Desktop can also run agents such as Codex CLI or Gemini CLI, billed by their provider. | | Where agents work | On the computer that runs OpenBot. | In virtual machines in Cognition's cloud, or on your computer with Devin Desktop or the Devin CLI. The agent's planning always runs in Cognition's cloud. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Devin Cloud keeps working when you close your laptop. Automations start sessions on a schedule or from events. | | Remote access | Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | A web app that you can install on your phone from the browser. Start sessions from Slack, Microsoft Teams, Linear, Jira or a GitHub comment. No iPhone or Android app is listed. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | You choose from Anthropic, OpenAI, Google, Cognition and open models, and the Adaptive router picks one for each task. Cognition describes no move when a provider is down. When your ChatGPT plan reaches its limit, a cloud session pauses. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | You save a username and password, site cookies or 2FA codes as encrypted Secrets, and Devin types them in its browser. You can take over its browser for CAPTCHAs and 2FA. Cognition names no password manager and no payment cards. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Voice calls in the web app, on a live speech model, and you can interrupt Devin while it talks. Cognition does not say where it processes speech. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Slack and Microsoft Teams: you tag @Devin to start a session. Cognition names no Discord, Telegram or WhatsApp. | | Teams | Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | Devin can hand parts of a task to managed Devins, each in its own virtual machine. The Teams plan takes up to 200 users, with no limit on parallel sessions. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | Code and sessions are processed in Cognition's cloud. By default, Cognition can use your data for model training; only paid plans can opt out. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Free with a light quota. Pro $20 a month, Max $200 a month, and Teams $80 a month plus $40 for each full seat. Every app needs a Devin account. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | Cognition lists no countries; its terms exclude countries under a US embargo. Devin works in Cognition's cloud, and Cognition names no EU data region. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | Devin Desktop and the Devin CLI for macOS, Windows and Linux, a web app, and plugins for JetBrains IDEs. No native mobile app is listed. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | Proprietary. Cognition's terms prohibit reverse engineering. | Devin is Cognition's autonomous AI software engineer: it can write, run and test code, and open the pull request. In 2025 Cognition acquired Windsurf, and in June 2026 Windsurf became Devin Desktop. Devin now has four parts: Devin Cloud, Devin Desktop, the Devin CLI and Devin Review. OpenBot's agents are coding agents too. The difference is where they run and whose plan pays for them. Devin works mainly in Cognition's cloud, paid through a Devin plan. OpenBot runs a team of agents on your own computer, with the AI plans you choose. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Devin Devin supports the latest models from Anthropic, OpenAI, Google and Cognition, and open-source models such as DeepSeek, Kimi and GLM. You choose a model in Devin Desktop and the CLI, and a mode in Devin Cloud. Cognition's own SWE-2 model is post-trained from Kimi K3. Devin does not take your own API keys: you pay through your Devin plan, and extra use is billed at API prices. On Pro, Max and Teams, Devin Desktop can also run third-party agents such as Codex CLI, Claude Agent, OpenCode and Gemini CLI; their provider bills you directly. ## Where the work happens ### OpenBot OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Devin Devin Cloud works in virtual machines in Cognition's cloud, and it keeps working when you close your laptop. Automations start sessions from a schedule, Slack, GitHub, Linear or a webhook. Devin Desktop and the Devin CLI run on your own computer. With Outposts, commands can run on machines that you operate. In every case, the agent's planning and inference run in Cognition's cloud; Enterprise can have a dedicated deployment. ## Phones and remote access ### OpenBot The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Devin Devin's web app can be installed on a phone from the browser, as a Progressive Web App. You can also start sessions from Slack, Microsoft Teams, Linear or Jira, from a GitHub pull request with a /devin comment, or through the Devin API. The sessions themselves run in Cognition's cloud. We found no Devin app in the App Store or Google Play. ## How agents work as a team ### OpenBot In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Devin Devin can give parts of a large task to a team of managed Devins that work in parallel, each in its own virtual machine, and the Devin CLI has subagents. Pro and Max run up to 10 sessions at the same time; Teams and Enterprise have no limit. The Teams plan takes up to 200 users, with shared work and an admin dashboard; Enterprise adds single sign-on and a dedicated deployment. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Devin Devin processes your code and sessions in Cognition's cloud on AWS, and keeps the data for as long as you are a customer. By default, Cognition can use your data for model training; on a paid plan you can opt out, and on Enterprise it trains only with your written consent. Cognition is SOC 2 Type II certified. A deployment in your own cloud is only for Enterprise. ## Questions ### Is OpenBot an alternative to Devin? Yes, if you want your coding agents on your own computer. Both give you agents that write, run and test code. OpenBot runs a team of agents on your computer with the AI plans you choose. Devin is an autonomous software engineer that works mainly in Cognition's cloud, paid through a Devin plan. ### Why choose OpenBot over Devin? Your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. OpenBot keeps your code, chats and files on your own computer, and OpenBot does not train models on them. It has native apps for iPhone and Android. The app is free, and its source code is on GitHub. Devin is the better fit when you want agents in the cloud that start from Slack, GitHub or Jira. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can I use my own API key or plan with Devin? Not for Devin's own agent: Cognition says Devin does not support third-party API keys, and you pay through your Devin plan. On Pro, Max and Teams, Devin Desktop can run third-party agents such as Codex CLI or Gemini CLI, and those providers bill you directly. ### Is Devin Desktop the same as Windsurf? Yes. Cognition acquired Windsurf in 2025, and in June 2026 it launched the next version of Windsurf as Devin Desktop. Its Devin Local agent replaces Cascade. Devin Local runs on your computer, but its model still runs in Cognition's cloud. ### Does Devin train on my code? It can. Cognition's terms say that it may use customer data for model training, and that customers on a paid plan can opt out. On Enterprise, Cognition trains only with your written consent. OpenBot does not train models; your provider's own terms apply to the requests you send it. ### Do agents keep working when my laptop is closed? Yes, with both. Devin Cloud works in Cognition's cloud, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### How much do OpenBot and Devin cost? OpenBot is free for noncommercial use, commercial use needs a separate license, and your agents use the plans you already pay for. Devin has a free plan with a light quota; Pro costs $20 a month, Max $200 a month, and Teams $80 a month plus $40 for each full seat. Extra use is billed at API prices. ## Sources Every statement about Devin comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Devin](https://devin.ai/) - [Devin pricing](https://devin.ai/pricing) - [Devin Cloud](https://devin.ai/cloud) - [Devin download](https://devin.ai/download) - [Windsurf is now Devin Desktop](https://devin.ai/blog/windsurf-is-now-devin-desktop) - [Windsurf's next chapter](https://devin.ai/blog/windsurfs-next-chapter) - [Devin Cloud in your terminal](https://devin.ai/blog/devin-cloud-in-your-terminal) - [Introducing Devin 2.2](https://cognition.com/blog/introducing-devin-2-2) - [SWE-2](https://cognition.com/blog/swe-2) - [Introducing Devin](https://docs.devin.ai/get-started/devin-intro) - [Devin CLI](https://docs.devin.ai/work-with-devin/devin-cli) - [Devin CLI models](https://docs.devin.ai/cli/models) - [Devin Desktop getting started](https://docs.devin.ai/desktop/getting-started) - [Devin Desktop third-party agents](https://docs.devin.ai/desktop/acp) - [Devin advanced capabilities](https://docs.devin.ai/work-with-devin/advanced-capabilities) - [Devin automations](https://docs.devin.ai/product-guides/automations) - [Devin Outposts](https://docs.devin.ai/cloud/outposts/overview) - [Devin integrations](https://docs.devin.ai/integrations/overview) - [Devin release notes](https://docs.devin.ai/release-notes/overview) - [Devin enterprise deployment](https://docs.devin.ai/enterprise/deployment/overview) - [Security at Cognition](https://docs.devin.ai/admin/security) - [Devin enterprise security](https://docs.devin.ai/enterprise/security-access/security/enterprise-security) - [Devin self-serve plans](https://docs.devin.ai/admin/billing/self-serve) - [Cognition platform terms](https://cognition.com/legal/platform-terms-of-service) - [Cognition Trust Center](https://trust.cognition.ai/) - [Devin Adaptive](https://docs.devin.ai/cli/adaptive) - [Devin with your ChatGPT plan](https://docs.devin.ai/admin/billing/chatgpt) - [Devin secrets and site cookies](https://docs.devin.ai/product-guides/secrets) - [Devin voice mode](https://docs.devin.ai/work-with-devin/voice-mode) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/claude-cowork [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Claude Cowork: A Local Alternative OpenBot vs Claude Cowork: run Claude Code, Codex, Gemini and Grok as a team on your own computer, with the AI plans you already pay for. Compare the two. Checked Oct 2, 2026 Sources ## The short answer Choose OpenBot if you want your agents on your own computer, with Claude and the other AI plans you already pay for, such as ChatGPT, Gemini or Grok: the app is free. Choose Claude Cowork if you use only Claude, and you want an agent in Anthropic's cloud that keeps working with no computer of yours on. ### Choose OpenBot if Recommended - You want Claude, ChatGPT, Gemini, Grok, Cursor, Cline, Muse Code and Pi agents in one team, or your own model. - Your files and chats must stay on your own computer, not in Anthropic's cloud. - You want a team of agents that give work to each other, each with its own job. - You want a free app, and source code that you can read. ### Choose Claude Cowork if - You use only Claude, and you already pay for a Claude plan. - You want an agent that works in the cloud, with no computer of yours to keep on. - You want to start and follow tasks from the Claude app on your phone or the web. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Claude Cowork:** Claude Cowork uses only Claude models. It needs a paid Claude plan: Pro from $17 a month, Max from $100, Team from $20 a seat, or Enterprise. ## At a glance OpenBot is better on 8 of 14 topics, and Claude Cowork on 0. OpenBot and Claude Cowork compared, as checked on Oct 2, 2026 | Topic | OpenBot | Claude Cowork | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Claude models only, with your paid Claude plan. Companies can also use Claude through Amazon Bedrock, Google Cloud or Microsoft Foundry. | | Where agents work | Better: On the computer that runs OpenBot. | In Anthropic's cloud. From 6 October 2026, new tasks on Pro and Max run only in the cloud; tasks already started on your computer stay there. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Cloud sessions keep working when you close your laptop, and scheduled tasks run with no device online. | | Remote access | Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | Start, steer and review tasks from the Claude apps for iPhone and Android or the web, in beta. Sessions and files are saved to your Claude account. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Claude models only. Claude changes the model only when a safety check flags a message. When you reach a limit, you wait or pay for usage credits. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | The built-in browser can import your saved logins, site by site. Claude in Chrome cannot make purchases or handle card data. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Cowork tasks take dictation, but not voice mode. In the new Claude app, voice mode works in any conversation, and Claude speaks back. Anthropic does not say where it processes speech. | | Chat apps | Better: Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Cowork has no chat app. Claude Tag, a separate Anthropic product for Team and Enterprise plans, works in Slack. | | Teams | Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | Claude can split a task across sub-agents that work in parallel. You cannot share a session, but you can share what it makes, and Team and Enterprise plans can share projects. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | Cloud sessions and their files are saved to your Claude account; local files that a cloud session opens are processed on Anthropic's servers. On Pro and Max, you can opt out of model training. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Needs a paid Claude plan: Pro from $17 a month, Max from $100, Team from $20 a seat, or Enterprise. Cowork uses your limits faster than chat. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | In the countries where Anthropic offers Claude, which include the US, the EU, the UK and Switzerland, but not China or Russia. The app is in 11 languages. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | The Claude desktop app for macOS and Windows, and Linux in beta. The Claude apps for iPhone and Android and the web, in beta. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | Closed source; the terms prohibit reverse engineering. Its plugins are on GitHub under the Apache License 2.0. | Claude Cowork is Anthropic's agent for work beyond code, built on the same agentic approach as Claude Code. It started in January 2026 as a research preview on the Mac, and on 16 September 2026 Anthropic started to roll it into Claude itself, first for Pro and Max. OpenBot also works with your Claude plan: Claude Code is one of its agents. The difference is where the agents run and how many providers you can use. Claude Cowork runs in Anthropic's cloud with Claude only. OpenBot runs a team of agents on your own computer, with Claude and the other AI plans you choose. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Claude Cowork Claude Cowork runs on Claude models, and it is available on the paid Claude plans: Pro, Max, Team and Enterprise. Companies can also use it through Amazon Bedrock, Google Cloud or Microsoft Foundry. Anthropic describes no way to use models from other companies. Cowork uses your plan's limits faster than chat. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Claude Cowork Claude Cowork sessions run in the cloud by default: the agent loop and the code run on Anthropic's servers, and sessions and files are saved to your Claude account. They keep working when you close your laptop, and scheduled tasks run with no device online. Existing desktop setups can still run sessions locally, in a Linux virtual machine on your computer, but from 6 October 2026, new tasks on Pro and Max run only in the cloud. On Team plans, cloud sessions are on by default; on Enterprise, an owner turns them on. ## Phones and remote access ### OpenBot The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Claude Cowork Claude Cowork runs on the web and in the Claude apps for iPhone and Android, in beta, on Pro, Max and Team plans, and on Enterprise plans where an admin turns it on. You start a task on one device, steer it from another, and pick up the result anywhere, because the session runs in Anthropic's cloud. ## How agents work as a team ### OpenBot In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Claude Cowork Claude Cowork is one agent that can break a complex task into smaller parts and give them to sub-agents that work at the same time. You cannot share a session with other people, but you can share the artifacts it makes. On Team and Enterprise plans, members can share projects, and owners can turn Cowork off for the organisation. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Claude Cowork Cloud sessions and their files are saved to your Claude account. Local files that a cloud session opens through the desktop app are processed on Anthropic's servers, not only on your computer. On Pro and Max, you can opt out of model training; if you allow it, Anthropic can keep your data for up to 5 years. A deleted conversation leaves Anthropic's storage within 30 days. ## Questions ### Is OpenBot an alternative to Claude Cowork? Yes, if you want your agents on your own computer. Both give you agents that do the work, and both work with your Claude plan. OpenBot runs a team of agents on your computer, with Claude Code and the other providers you choose. Claude Cowork is Anthropic's agent in its cloud, with Claude only. ### Why choose OpenBot over Claude Cowork? Your agents can use Claude, ChatGPT, Gemini or Grok, or your own model, and they work together as a team. OpenBot keeps your workspaces, chats and files on your own computer. The app is free, and its source code is on GitHub. Claude Cowork is the better fit when you use only Claude and want an agent with no computer of yours to keep on. ### Can I use my Claude plan with OpenBot? Yes. OpenBot runs Claude Code with your own Claude plan, so the Claude plan that you pay for works in OpenBot too. Next to it, Codex uses your ChatGPT plan, Gemini a Google AI Pro or Ultra plan, Grok CLI your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Is Claude Cowork still a separate product? It is becoming part of Claude. On 16 September 2026, Anthropic started to merge Cowork and chat into one Claude, first on Pro and Max, with more plans to follow. You start a Cowork task from the same message box as a chat. ### Does Claude Cowork run on my computer? By default, no: sessions run in Anthropic's cloud, and local files that a session opens are processed on Anthropic's servers. From 6 October 2026, new tasks on Pro and Max run only in the cloud; tasks already started in a virtual machine on your computer stay there. ### Is Claude Cowork free? No. Claude Cowork needs a paid Claude plan: Pro costs $17 a month with annual billing or $20 monthly, Max $100 or $200 a month, and Team $20 or $25 a seat for standard seats. OpenBot is free for noncommercial use, and it uses the plans you already pay for. ### Do agents keep working when my laptop is closed? Yes, with both. Claude Cowork works in Anthropic's cloud, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### Is Claude Cowork open source? No. Claude Cowork is closed source, and Anthropic's terms prohibit reverse engineering. Its plugins are on GitHub under the Apache License 2.0. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0. ## Sources Every statement about Claude Cowork comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Claude Cowork](https://claude.com/product/cowork) - [Claude Cowork and chat are now one Claude](https://claude.com/blog/cowork-is-now-claude) - [Claude Cowork for enterprise](https://claude.com/blog/cowork-for-enterprise) - [Get started with Claude Cowork](https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork) - [Claude Cowork architecture](https://support.claude.com/en/articles/14479288-claude-cowork-architecture-overview) - [Use Claude Cowork safely](https://support.claude.com/en/articles/13364135-use-claude-cowork-safely) - [Claude Cowork on web, desktop and mobile](https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile) - [Claude Cowork on Team and Enterprise](https://support.claude.com/en/articles/13455879-use-claude-cowork-on-team-and-enterprise-plans) - [Claude Cowork and chat are one Claude](https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude) - [Scheduled tasks in Claude Cowork](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork) - [Projects in Claude Cowork](https://support.claude.com/en/articles/14116274-organize-your-tasks-with-projects-in-claude-cowork) - [Install Claude Desktop](https://support.claude.com/en/articles/10065433-install-claude-desktop) - [Claude release notes](https://support.claude.com/en/articles/12138966-release-notes) - [Countries where Claude is offered](https://www.anthropic.com/supported-countries) - [Claude in your language](https://support.claude.com/en/articles/10769299-using-claude-in-your-preferred-language) - [Claude pricing](https://claude.com/pricing) - [Claude download](https://claude.com/download) - [How long Anthropic stores your data](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data) - [Is my data used for model training?](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training) - [Anthropic consumer terms](https://www.anthropic.com/legal/consumer-terms) - [Claude knowledge work plugins](https://github.com/anthropics/knowledge-work-plugins) - [Why Claude switched models](https://support.claude.com/en/articles/17161993-why-claude-switched-models-in-your-conversation-with-sonnet-5-5) - [Claude usage credits](https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans) - [The built-in browser in Claude Cowork](https://support.claude.com/en/articles/16607400-use-the-built-in-browser-in-claude-cowork) - [Claude in Chrome permissions](https://support.claude.com/en/articles/12902446-claude-in-chrome-permissions-guide) - [Claude voice mode](https://support.claude.com/en/articles/11101966-use-voice-mode) - [Claude Tag overview](https://claude.com/docs/claude-tag/overview) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/manus [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Manus: A Local Manus Alternative OpenBot is a free, local Manus alternative: run a team of AI agents on your own computer with the ChatGPT, Claude, Gemini or Grok plan you already pay for. Checked Oct 2, 2026 Sources ## The short answer Choose OpenBot if you want your agents on your own computer, with the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok: the app is free. Choose Manus if you want a general agent that works in its own cloud computer, keeps working with no computer of yours on, and chooses the model for you. ### Choose OpenBot if Recommended - You already pay for ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code, or you run your own model. - Your files and chats must stay on your own computer, not in a cloud in the US or Singapore. - You want a team of coding agents that give work to each other, each with its own job. - You want a desktop app for Linux, or a free app. ### Choose Manus if - You want an agent that works in a cloud computer, with no computer of yours to keep on. - You want research, websites and slides from one prompt, with many agents in parallel. - You want to give it tasks in LINE or by email. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Manus:** Manus chooses the model for each task; you choose only between Manus 2.0 Lite, Manus 2.0 and Manus 2.0 Max. You pay Manus in credits: 300 free credits a day, or Pro plans from $20 a month. ## At a glance OpenBot is better on 8 of 14 topics, and Manus on 0. OpenBot and Manus compared, as checked on Oct 2, 2026 | Topic | OpenBot | Manus | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Manus chooses the model for each task. You choose Manus 2.0 Lite, Manus 2.0 or Manus 2.0 Max. Manus states no way to use your own AI plan, API key or model. | | Where agents work | Better: On the computer that runs OpenBot. | In a cloud sandbox that Manus runs, by default. With the desktop app, Manus can also run commands in the folders that you allow on your Mac or Windows PC. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Cloud tasks keep working after you close the app, and scheduled tasks run while you are offline. Tasks on your own computer need it on. | | Remote access | Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | Apps for iPhone and Android and a web app. You can also give tasks in Telegram, WhatsApp Business, LINE, Slack or email. The tasks and their files stay in Manus's cloud. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Manus chooses the model for each task. With Manus Flex, you can use your own API key. Manus states no move to a backup model. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | You sign in to sites yourself in Manus's cloud browser, and Manus keeps the session but not your password. You take over the browser for codes and CAPTCHAs. Manus states no password manager or saved cards. | | Voice | Better: Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | You can give a voice command in the phone app, and Manus writes out the voice messages that you send in chat apps. Manus states no reading of replies aloud. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Slack, Telegram, LINE and email. Cue, its app for personal agents, also connects to iMessage. | | Teams | Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | Wide Research runs many agents in parallel on one task. Up to 50 people can work in one task, and the Team plan pools credits across seats. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. Product analytics, with no chat content, are on by default and linked to your account when you sign in; you can turn them off. | Tasks, files and the commands the agent runs are stored in Manus's cloud, in the US and Singapore. Its terms give Manus a permanent license to use your content, in aggregate, to improve the service. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Free with 300 credits a day. Pro plans from $20 to $200 a month, and the Team plan from $20 a seat. Needs an account, and is for people aged 18 or more. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | Worldwide, except countries under a US embargo, for people 18 and over. Its data is stored in the US and Singapore. The website is in 17 languages. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | macOS and Windows, apps for iPhone and Android, and the web. A Linux desktop app: not stated by Manus. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | Closed source. Its terms prohibit reverse engineering. | Manus is a general AI agent from Butterfly Effect, a company in Singapore. Meta acquired Manus in December 2025, and Manus resumed independent operations on 1 September 2026. Both products give you agents that do the work, not only answers. The difference is where the agents run and which model does the thinking. Manus works in its own cloud computer and chooses the model for you. OpenBot runs a team of agents on your own computer, with the AI plans you choose. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Manus Manus says it is LLM-agnostic: it chooses the right model for each task, so you do not have to. It does not name the models. You choose only between its own tiers: the free plan gets Manus 2.0 Lite in Agent mode, and paid plans also get Manus 2.0 and Manus 2.0 Max. You pay Manus in credits, and a plan that you already pay for at another AI company does not apply. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Manus By default, each Manus task works in a temporary sandbox: a virtual computer with internet access in Manus's cloud. A Cloud Computer that stays on costs $30 or $50 a month more. Cloud tasks and scheduled tasks keep running while you are offline. Since March 2026, the desktop app for macOS and Windows can also let Manus run commands in the folders that you allow on your own computer; those tasks need that computer on. ## Phones and remote access ### OpenBot The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Manus Manus has apps for iPhone and Android and a web app. You can also give it tasks in Telegram, WhatsApp Business, LINE or Slack, or by email; one messaging app at a time for each workspace. The work itself happens in Manus's cloud, or on your computer when you use the desktop app. ## How agents work as a team ### OpenBot In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Manus Manus is one agent that can split a large task: Wide Research starts hundreds of independent agents that work in parallel. Paid plans run up to 20 tasks at the same time. Up to 50 people can work in one task, and only the owner of the task pays the credits. The Team plan pools credits and adds single sign-on; the owner of a team can see all the session data of its members. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### Manus Manus stores your data in the US and Singapore. Its privacy policy says that it collects the files of each task, the shell commands the agent runs and their output, and the code. Manus says its model providers do not train on your data. For individual plans, it states no training opt-out, and its terms give Manus a permanent license to use your content, in aggregate, to improve the service. The Team plan says Manus does not train models on your data. ## Questions ### Is OpenBot an alternative to Manus? Yes, if you want your agents on your own computer. Both give you AI agents that do the work. OpenBot runs a team of agents on your computer with the AI plans you choose. Manus is a general agent that works in a cloud computer that Manus runs, with models that Manus chooses. ### Why choose OpenBot over Manus? Your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. OpenBot keeps your workspaces, chats and files on your own computer, it runs a team of coding agents, and it has desktop apps for macOS, Windows and Linux. The app is free, and its source code is on GitHub. Manus is the better fit when you want an agent with no computer of yours to keep on. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can I use my own AI plan or model with Manus? Manus states no way to do this. It chooses the model for each task, and you pay Manus in credits. You can choose only between Manus 2.0 Lite, Manus 2.0 and Manus 2.0 Max. ### Is Manus owned by Meta? Not now. Meta acquired Manus in December 2025. In August 2026, Manus said it was separating from Meta to comply with regulatory requirements, and it resumed independent operations on 1 September 2026, led by its founding team. ### Is OpenManus the same as Manus? No. OpenManus is a separate open-source project under the MIT License. Manus itself is closed source, and its terms prohibit reverse engineering. ### Does Manus work on my computer? Partly. By default, Manus works in a cloud sandbox. The desktop app for macOS and Windows can let it run commands in the folders that you allow on your computer, while the app runs. Your tasks and their data are still stored in Manus's cloud. ### Do agents keep working when my laptop is closed? Yes, with both. Manus works in its cloud, so no computer of yours must stay on. OpenBot agents work on the computer or server that runs OpenBot: keep that computer on, or use a hosted OpenBot server, and connect from your laptop or phone. ### How much do OpenBot and Manus cost? OpenBot is free for noncommercial use, commercial use needs a separate license, and your agents use the plans you already pay for. Manus is free with 300 credits a day; Pro plans cost $20 to $200 a month and the Team plan starts at $20 a seat. Manus payments are not refundable. ## Sources Every statement about Manus comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Manus](https://manus.im/) - [Introducing Manus 2.0](https://manus.im/blog/introducing-manus-2-0) - [Manus pricing](https://manus.im/pricing) - [Manus membership pricing](https://help.manus.im/en/articles/11711111-what-is-the-current-membership-pricing-for-manus) - [Manus Team plan](https://manus.im/team) - [Manus: A note to our users](https://manus.im/blog/a-note-to-our-users) - [Manus resumes independent operations](https://manus.im/blog/manus-resumes-independent-operations) - [Manus joins Meta](https://manus.im/blog/manus-joins-meta-for-next-era-of-innovation) - [Manus docs: Welcome](https://manus.im/docs/introduction/welcome) - [Manus docs: Desktop](https://manus.im/docs/features/desktop) - [Manus docs: Wide Research](https://manus.im/docs/features/wide-research) - [Manus docs: Scheduled tasks](https://manus.im/docs/features/scheduled-tasks) - [Manus My Computer](https://manus.im/blog/manus-my-computer-desktop) - [Manus desktop app download](https://help.manus.im/en/articles/14089011-how-to-download-the-manus-desktop-app) - [Manus Cloud Computer](https://help.manus.im/en/articles/15392111-what-is-the-cloud-computer) - [Manus Cloud Computer plans](https://help.manus.im/en/articles/15392078-understanding-cloud-computer-plans-and-billing) - [Manus Collab](https://help.manus.im/en/articles/12135428-how-can-i-use-manus-collab) - [Manus in Telegram](https://manus.im/blog/manus-agents-telegram) - [One messaging agent at a time](https://help.manus.im/en/articles/14178640-can-i-connect-multiple-messaging-agents-at-once) - [Mail Manus](https://help.manus.im/en/articles/12059745-how-can-i-use-mail-manus) - [Manus Trust Center FAQ](https://trust.manus.im/faq) - [Manus privacy policy](https://manus.im/privacy) - [Manus terms of use](https://manus.im/terms) - [Manus download](https://manus.im/download) - [OpenManus on GitHub](https://github.com/FoundationAgents/OpenManus) - [Introducing Manus Flex](https://manus.im/blog/introducing-manus-flex) - [Manus Cloud Browser](https://manus.im/docs/features/cloud-browser) - [What is new in Manus 2.0](https://help.manus.im/en/articles/17190150-what-is-new-in-manus-2-0) - [Manus on LINE](https://help.manus.im/en/articles/14495248-sending-and-receiving-messages-on-line) - [Manus in Slack](https://help.manus.im/en/articles/14432468-how-do-i-set-up-connect-and-use-manus-agents-in-slack) - [Cue messaging apps](https://help.manus.im/en/articles/14178640-can-i-connect-multiple-messaging-agents-at-once-in-cue) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/openclaw [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs OpenClaw: An OpenClaw Alternative OpenBot vs OpenClaw (formerly Clawdbot and Moltbot): run Codex, Claude Code, Gemini and Grok as a team on your computer, and reach them from your phone. Checked Oct 2, 2026 Sources ## The short answer OpenBot and OpenClaw are close: both run AI agents on your own computer, use the AI plans you already pay for, and have apps for desktop, iPhone and Android. Choose OpenBot if you want Codex, Claude Code, Gemini and Grok CLI to work as a team of separate agents, and to reach them from your phone anywhere with no VPN. Choose OpenClaw if you want one personal assistant that talks to you in WhatsApp, Signal or iMessage, under the MIT license. ### Choose OpenBot if Recommended - You want Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, Pi and Muse Code as separate agents in one team, each with its own job. - You want your phone to reach your agents from anywhere, with no VPN, tunnel or public address to set up. - You pay for Google AI Pro or Ultra and want your agents to use that plan. - You want your agents to work 24/7 with no VPS to set up: use a hosted OpenBot server. ### Choose OpenClaw if - You want to talk to your assistant in WhatsApp, Signal or iMessage. - You need the MIT license, for example for commercial use. - You want many plugins and skills from a large open-source community. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **OpenClaw:** OpenClaw signs in with ChatGPT or Codex, a Claude Code login, SuperGrok or X Premium, and GitHub Copilot, or uses API keys and local models. Gemini needs an API key or Vertex AI: OpenClaw does not offer a new Google sign-in. ## At a glance OpenBot is better on 4 of 14 topics, and OpenClaw on 2. OpenBot and OpenClaw compared, as checked on Oct 2, 2026 | Topic | OpenBot | OpenClaw | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Sign in with ChatGPT, a Claude Code login, SuperGrok or X Premium, or GitHub Copilot. Or use API keys and local models. Gemini needs an API key or Vertex AI; a consumer Gemini plan does not work. | | Where agents work | On the computer that runs OpenBot. | On the computer that runs the OpenClaw Gateway: your own computer, a home server or a VPS that you rent. | | When your computer is off | Better: Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Scheduled jobs run on the computer that runs the Gateway. On a laptop, sleep and restarts stop it. OpenClaw has no hosted tier: for 24/7 work it recommends a VPS that you rent, set up and update over SSH. | | Remote access | Better: Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | Apps for iPhone and Android connect to the Gateway over your local network, a tailnet such as Tailscale, an SSH tunnel, or a public HTTPS address that you set up. Messaging apps such as WhatsApp and Telegram also work, and carry your messages. | | Backup models | When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | After rate limits, billing errors and timeouts, OpenClaw tries your other accounts at the same provider, then the backup models that you list. A model that you chose by hand for a session does not change. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | OpenClaw recommends that you sign in to sites yourself in its browser. On a Mac, it can sign in with 1Password: you confirm each use, and the password never reaches the model. OpenClaw states no payment cards. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Talk mode on Mac, iPhone and Android listens, answers aloud, and stops when you speak. A local model can read replies on a Mac, but speech recognition can use Apple's online service. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Better: WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams, Matrix and more than 20 others. | | Teams | Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | Several isolated agents in one Gateway, each with its own workspace, and subagents for parallel work. Agents can run through Codex, Claude Code or Gemini CLI. A team can share one Gateway, with shared sessions and roles. | | Your data | Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. | Sessions, memory and settings stay in ~/.openclaw on the computer that runs the Gateway. | | Price and account | Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Better: Free under the MIT license, also for commercial use. No subscription and no hosted tier. You pay your model provider, and a VPS if you rent one. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | Anywhere: you run it yourself, and there is no hosted version. The Control UI is in 21 languages. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | macOS, Windows and Linux, apps for iPhone and Android, a web Control UI, and a command line. | | Source code | On GitHub, under the PolyForm Noncommercial License 1.0.0. | On GitHub, under the MIT License. | OpenClaw started as Warelay, a WhatsApp gateway, then was called Clawdbot and Moltbot, and took its current name in January 2026. Like OpenBot, it runs AI agents on a computer that you choose, keeps your data there, works with the AI plans you pay for, and has apps for desktop and phone. The difference is how the agents are organised and how you reach them. In OpenBot, each provider's coding tool is its own agent in a team, and your phone connects from anywhere. OpenClaw is a personal assistant that runs in a Gateway: it can run its turns through Codex, Claude Code or Gemini CLI, and it talks to you through messaging apps. ## Models and the plans you pay for ### OpenBot OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### OpenClaw OpenClaw signs in with a ChatGPT account or a Codex login, reuses a Claude Code login on the same computer, and recommends Grok sign-in with a SuperGrok or X Premium subscription. It also works with GitHub Copilot, API keys from many providers, and local models through Ollama, LM Studio, vLLM or llama.cpp. Gemini needs a Google API key or Vertex AI: OpenClaw does not offer a new Gemini CLI or Antigravity sign-in, because Google ended that sign-in for consumer Gemini CLI and the Antigravity terms prohibit third-party tools. ## Where the work happens ### OpenBot OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### OpenClaw OpenClaw runs in a Gateway on the computer that you choose. The desktop apps install the Gateway for you; you can also run it on a home server, in Docker, or on a VPS. Scheduled jobs and a regular heartbeat run in the Gateway. On a laptop, sleep, network drops and restarts disconnect it, so OpenClaw recommends a VPS for 24/7 work. ## Phones and remote access ### OpenBot Better The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### OpenClaw The OpenClaw apps for iPhone and Android connect to the Gateway over your local network or a tailnet. The Gateway listens only on the computer itself by default; to reach it from outside, OpenClaw recommends Tailscale, an SSH tunnel, or a public HTTPS address that you set up. You can also talk to OpenClaw in WhatsApp, Telegram, Slack, Discord, Signal, iMessage and many more; your messages then go through the messaging service that you choose. ## How agents work as a team ### OpenBot In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### OpenClaw One OpenClaw Gateway can run several isolated agents, each with its own workspace and session history, and messages go to the right agent by rules. An agent can run its turns through Codex, Claude Code or Gemini CLI, and it can start subagents, 5 active in each session by default. Through ACP, it can also give work to Claude Code, Codex, Cursor, Gemini CLI and others. A team can share one Gateway, with shared sessions and roles; OpenClaw says a Gateway is one trust domain, for people who already trust each other. ## Data and privacy ### OpenBot OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds your profile, team memberships, invitations, sign-in sessions, the settings that let devices find each other, and any agent templates that you publish. Product analytics, which never include chat content, are on by default and linked to your account when you sign in; you can turn them off in Settings. ### OpenClaw OpenClaw keeps settings, sessions and memory under ~/.openclaw on the computer that runs the Gateway, with session history in SQLite and memory in Markdown files. A daily update check is on by default and sends the app version and system details; anonymous feature statistics are off until you turn them on. OpenClaw says that third-party skills and inbound messages are untrusted input. ## Questions ### Is OpenBot an alternative to OpenClaw? Yes. Both run AI agents on your own computer and keep your data there. OpenBot runs the coding tools of the providers you pay for, such as Codex and Claude Code, as a team, and your phone reaches them from anywhere. OpenClaw is a personal assistant that runs in a Gateway and talks to you in messaging apps. ### Is OpenClaw the same as Clawdbot or ClawBot? OpenClaw started as Warelay and was then called Clawdbot. It became Moltbot in January 2026 and OpenClaw a few days later. ClawBot by Ipsion AI is a different, hosted product, and it says it is not part of the OpenClaw project. ### What is the main difference between OpenBot and OpenClaw? How the agents are organised and how you reach them. In OpenBot, each provider's coding tool, such as Codex with your ChatGPT plan or Claude Code with your Claude plan, is its own agent, and the agents work as a team. OpenClaw is one assistant that can run its turns through Codex, Claude Code or Gemini CLI, and you reach it through messaging apps or its own apps on your network. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can I use my Claude or Gemini subscription with OpenClaw? Claude, yes: OpenClaw can reuse a Claude Code login on the same computer, and that use comes from your plan's limits. Gemini, no: OpenClaw needs a Google API key or Vertex AI, and it does not offer a new Google account sign-in. ### Can I reach my agents from my phone? Yes, with both. The OpenBot apps connect from anywhere over an encrypted connection between your devices, with no VPN. The OpenClaw apps connect over your local network, a tailnet, an SSH tunnel or a public HTTPS address that you set up, and you can also talk to OpenClaw in messaging apps such as WhatsApp or Telegram. ### Is OpenClaw open source? Yes. OpenClaw is on GitHub under the MIT License, which also allows commercial use. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0, and commercial use needs a separate license. ### Do my chats leave my computer? With both, your chats stay on the computer that runs the app, and the model provider that you choose gets the requests. With OpenClaw, a messaging app that you connect also carries those messages. ### Do agents keep working when my laptop is closed? Yes, with both, when they run on a computer that stays on. With OpenBot, run it on a desktop, a server or a hosted OpenBot server, and connect from your laptop or phone. OpenClaw recommends a VPS or a home server for its Gateway. ### How much do OpenBot and OpenClaw cost? Both apps are free. OpenBot is free for noncommercial use, and commercial use needs a separate license. OpenClaw is free under the MIT License, with no subscription and no hosted tier. With both, you pay your model provider. ## Sources Every statement about OpenClaw comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [OpenClaw on GitHub](https://github.com/openclaw/openclaw) - [OpenClaw](https://openclaw.ai/) - [Introducing OpenClaw](https://openclaw.ai/blog/introducing-openclaw) - [OpenClaw lore](https://docs.openclaw.ai/start/lore) - [OpenClaw Anthropic provider](https://docs.openclaw.ai/providers/anthropic) - [OpenClaw Google provider](https://docs.openclaw.ai/providers/google) - [OpenClaw xAI provider](https://docs.openclaw.ai/providers/xai) - [OpenClaw OpenAI authentication](https://docs.openclaw.ai/providers/openai/authentication) - [OpenClaw iOS app](https://docs.openclaw.ai/platforms/ios) - [OpenClaw Android app](https://docs.openclaw.ai/platforms/android) - [OpenClaw remote access](https://docs.openclaw.ai/gateway/remote) - [OpenClaw agent runtimes](https://docs.openclaw.ai/concepts/agent-runtimes) - [OpenClaw ACP agents](https://docs.openclaw.ai/tools/acp-agents) - [OpenClaw subagents](https://docs.openclaw.ai/tools/subagents/operations) - [OpenClaw channels](https://docs.openclaw.ai/channels) - [OpenClaw scheduled jobs](https://docs.openclaw.ai/automation/cron-jobs) - [OpenClaw files on disk](https://docs.openclaw.ai/help/faq/where-things-live-on-disk) - [OpenClaw multi-agent routing](https://docs.openclaw.ai/concepts/multi-agent) - [OpenClaw for teams](https://docs.openclaw.ai/start/teams) - [OpenClaw telemetry](https://docs.openclaw.ai/gateway/telemetry) - [OpenClaw hosting FAQ](https://docs.openclaw.ai/help/faq-first-run/providers-and-hosting) - [ClawBot by Ipsion AI](https://clawbot-ai.app/) - [OpenClaw Control UI settings](https://docs.openclaw.ai/web/control-ui/settings) - [OpenClaw model failover](https://docs.openclaw.ai/concepts/model-failover) - [OpenClaw browser login](https://docs.openclaw.ai/tools/browser-login) - [OpenClaw 1Password](https://docs.openclaw.ai/gateway/1password) - [OpenClaw Talk mode](https://docs.openclaw.ai/nodes/talk) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/hermes-agent [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Hermes Agent: A Hermes Agent Alternative OpenBot vs Hermes Agent: run Codex, Claude Code, Gemini and Grok as a team on your computer, with the AI plans you already pay for. Compare the two. Checked Oct 2, 2026 Sources ## The short answer Both apps run AI agents on your own computer. Choose OpenBot if you want Codex, Claude Code, Gemini and Grok CLI to work as a team with the plans you already pay for, including Claude Pro or Max and Google AI Pro, and to reach them from iPhone and Android apps. Choose Hermes Agent if you want an open-source agent under the MIT license that learns skills over time and talks to you in messaging apps such as WhatsApp or Signal. ### Choose OpenBot if Recommended - You pay for Claude Pro or Max, or Google AI Pro or Ultra, and want your agents to use that plan. - You want Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, Pi and Muse Code as agents in one team, each with its own job. - You want iPhone and Android apps that connect to your own computer. - Coworkers must join your agents on one team host, with shared channels and files. ### Choose Hermes Agent if - You want an agent that makes its own skills and searches its past conversations. - You want to talk to it in WhatsApp, Signal or email. - You need the MIT license, for example for commercial use. - You want Nous Research to host it for you in Hermes Cloud. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Hermes Agent:** Hermes Agent uses API keys from many providers, and signs in with ChatGPT, SuperGrok or GitHub Copilot. As its model, a Claude plan works only on Max with extra usage credits, and a consumer Gemini plan does not work. ## At a glance OpenBot is better on 3 of 14 topics, and Hermes Agent on 2. OpenBot and Hermes Agent compared, as checked on Oct 2, 2026 | Topic | OpenBot | Hermes Agent | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | API keys from many providers, OpenRouter, and local models. Sign in with ChatGPT, SuperGrok or GitHub Copilot. As its model, a Claude plan works only on Max with extra usage credits; a consumer Gemini plan does not work. | | Where agents work | On the computer that runs OpenBot. | On your computer, a server, or in Docker. Hermes Cloud, which Nous Research hosts, costs a daily price plus the models you use. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Scheduled tasks run on the server or cloud computer that runs Hermes. On a laptop, they stop while Hermes is stopped. | | Remote access | Better: Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | Messaging apps such as Telegram, WhatsApp, Signal and Slack, which carry your messages. No native iPhone or Android app is stated. | | Backup models | When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | When a provider has rate limits or server errors, Hermes Agent moves to backup models from other providers by itself, in the order that you set. | | Logins and payments | The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | An encrypted vault on your computer keeps logins, payment cards and addresses. Passwords and card numbers go only into the page, and Hermes asks you before each card fill. It also uses 1Password or Bitwarden. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Push-to-talk in the terminal, and spoken replies in Telegram and Discord. Speech recognition can run on your computer with no API key. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Better: Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, Matrix and about 20 more. | | Teams | Better: Each agent is a full coding agent with its own job. They give work to each other in shared channels. Coworkers join one team host, and roles set who can view, chat with or manage each agent. | 10 subagents at once by default, and you can set more. Bots with their own roles talk in group chats of up to six. Several people can use one Hermes instance through messaging apps, with an allowlist. | | Your data | Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. | Conversations, memory and skills stay in ~/.hermes on your computer, with no telemetry. The provider you choose gets the requests you send. | | Price and account | Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Better: Free under the MIT license, also for commercial use. No account is necessary. You pay your model provider, and Hermes Cloud if you use it. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | Anywhere: you run it yourself. Hermes Cloud uses the nearest of 14 locations in Europe, North America, Asia and Australia. The interface is in 17 languages. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | A command line, a terminal UI, a web dashboard, and Hermes Desktop for macOS, Windows and Linux. The command line also runs on Android in Termux. No iPhone or Android app is stated. | | Source code | On GitHub, under the PolyForm Noncommercial License 1.0.0. | On GitHub, under the MIT License. | OpenBot and Hermes Agent are close: both run AI agents on the computer that you choose, keep your chats there, and publish their source code. The difference is what an agent is. In OpenBot, each agent is a coding tool from the provider whose plan you pay for, and the agents work as a team. Hermes Agent is its own agent that calls a model through an API, learns skills as it works, and can run several named Bots. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Hermes Agent Hermes Agent calls a model through an API. It works with many providers, OpenRouter and Nous Portal, and with local models on any OpenAI-compatible server, such as Ollama or vLLM. It can sign in with a ChatGPT account, a SuperGrok or X Premium+ subscription, or GitHub Copilot. As its own model, a Claude plan works only on Max, and then it uses only the extra usage credits that you buy; with Claude Pro, you need an API key. There is no way to use a consumer Gemini plan: Gemini needs a Google API key. ## Where the work happens ### OpenBot OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Hermes Agent Hermes Agent runs on your computer, on a server, or in Docker. Its commands can run on that computer, in Docker, over SSH, or in cloud sandboxes such as Modal and Daytona. Hermes Cloud runs it on a computer that Nous Research hosts, for a daily price, and the models and tools it uses cost extra. On a laptop, tasks stop while Hermes is stopped. ## Phones and remote access ### OpenBot Better The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Hermes Agent Hermes Agent reaches your phone through messaging apps: Telegram, WhatsApp, Signal, Slack, SMS, email, iMessage through BlueBubbles, and many more. Your messages then go through the messaging service that you choose. Hermes Desktop can also connect to a Hermes backend on another computer. The documentation states no native iPhone or Android app. ## How agents work as a team ### OpenBot Better In OpenBot, each agent is a full coding agent: Codex, Claude Code, Gemini, Grok CLI, Cursor CLI, Cline, OpenCode, Pi, Muse Code or an agent of your own, each with its own job and workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Hermes Agent Hermes Agent starts up to 10 subagents at once by default, and you can set a higher limit. In Bot Mode, each Bot has its own role, model, memory and skills; Bots talk in group chats of up to six and send messages to each other. Bundled skills can give coding work to Claude Code, Codex or OpenCode. Several people can use one Hermes instance through messaging apps such as Slack or Telegram, and an allowlist controls who can talk to it. ## Data and privacy ### OpenBot OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds only your profile, team memberships, invitations, sign-in sessions and the settings that let devices find each other. ### Hermes Agent Hermes Agent keeps conversations, memory and skills in ~/.hermes on the computer that runs it, with its sessions in SQLite. Nous Research says Hermes Agent collects no telemetry, usage data or analytics, and API calls go only to the provider that you set up. ## Questions ### Is OpenBot an alternative to Hermes Agent? Yes. Both run AI agents on your own computer and keep your chats there. OpenBot runs the coding tools of the providers you pay for, such as Codex and Claude Code, as a team. Hermes Agent is its own agent loop that calls a model through an API. ### What is the main difference between OpenBot and Hermes Agent? What an agent is. An OpenBot agent is a provider's own coding tool, such as Codex with your ChatGPT plan or Claude Code with your Claude plan, and several agents work as a team. Hermes Agent is its own agent loop over many model providers. It learns skills as it works, and it talks to you in messaging apps. ### Can I use my Claude subscription with Hermes Agent? As the model of Hermes Agent itself, only on Claude Max, and then it uses only the extra usage credits that you buy, not your plan's own usage. With Claude Pro, you need an API key. Its bundled Claude Code skill can also give coding work to Claude Code with your Pro or Max plan. In OpenBot, Claude Code is the agent itself and uses your Claude plan. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can I use local models? Yes, with both. In OpenBot, an agent can use any OpenAI-compatible server through OpenCode, also a model on your own computer. Hermes Agent works with any OpenAI-compatible server, such as Ollama, vLLM or llama.cpp. ### Is Hermes Agent open source? Yes. Hermes Agent is on GitHub under the MIT License, which also allows commercial use. OpenBot's source code is on GitHub under the PolyForm Noncommercial License 1.0.0, and commercial use needs a separate license. ### Does Hermes Agent have a mobile app? The documentation states no native iPhone or Android app. You reach Hermes Agent from your phone through messaging apps such as Telegram, WhatsApp or Signal. OpenBot has apps for iPhone and Android that connect to your own computer over an encrypted connection. ### Do my chats leave my computer? With both, your chats stay on the computer that runs the app, and the model provider that you choose gets the requests. With Hermes Agent, a messaging app that you connect also carries those messages. ### Do agents keep working when my laptop is closed? Yes, with both, when they run on a computer that stays on. With OpenBot, run it on a desktop, a server or a hosted OpenBot server, and connect from your laptop or phone. Hermes Agent can run on a server, or in Hermes Cloud for a daily price. ### How much do OpenBot and Hermes Agent cost? Both apps are free. OpenBot is free for noncommercial use, and commercial use needs a separate license. Hermes Agent is free under the MIT License. With both, you pay your model provider. Hermes Cloud costs from $0.56 a day for a running Medium instance, plus the models and tools it uses. ## Sources Every statement about Hermes Agent comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Hermes Agent on GitHub](https://github.com/NousResearch/hermes-agent) - [Hermes Agent AI providers](https://hermes-agent.nousresearch.com/docs/integrations/providers) - [Hermes Agent FAQ](https://hermes-agent.nousresearch.com/docs/reference/faq) - [Hermes Agent platform support](https://hermes-agent.nousresearch.com/docs/getting-started/platform-support) - [Hermes Desktop](https://hermes-agent.nousresearch.com/docs/user-guide/desktop) - [Hermes Agent messaging](https://hermes-agent.nousresearch.com/docs/user-guide/messaging/) - [Hermes Agent subagents](https://hermes-agent.nousresearch.com/docs/user-guide/features/delegation) - [Hermes Agent Bot Mode](https://hermes-agent.nousresearch.com/docs/user-guide/bot-mode) - [Hermes Agent Claude Code skill](https://hermes-agent.nousresearch.com/docs/user-guide/skills/bundled/autonomous-ai-agents/autonomous-ai-agents-claude-code) - [Hermes Cloud](https://portal.nousresearch.com/cloud) - [Hermes Agent language packs](https://hermes-agent.nousresearch.com/docs/user-guide/features/language-packs) - [Hermes Agent fallback providers](https://hermes-agent.nousresearch.com/docs/user-guide/features/fallback-providers) - [Hermes Agent credential vault](https://hermes-agent.nousresearch.com/docs/user-guide/features/credential-vault) - [Hermes Agent voice mode](https://hermes-agent.nousresearch.com/docs/user-guide/features/voice-mode) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/muse [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Muse: A Local Alternative to Meta's Muse OpenBot is a free, local alternative to Meta's Muse: run Muse Code, Codex, Claude Code and Gemini as a team on your computer, with the plans you have. Checked Oct 2, 2026 Sources ## The short answer For most people who want agents for their work, OpenBot is the better choice: your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. They run on your own computer, and the app is free. Choose Muse if you want one personal assistant from Meta that runs on Meta's cloud and helps with email and purchases. ### Choose OpenBot if Recommended - You already pay for ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code, or you run your own model. - Your files and chats must stay on your own computer, not on Meta's cloud. - You want a team of agents that give work to each other, not one assistant. - You live outside the US and Canada, or you want a desktop app for Windows or Linux. - You want a free app, and source code that you can read. ### Choose Muse if - You want a personal assistant for email and purchases, and no computer to keep on. - You want to talk to it in WhatsApp, and later through Meta's AI glasses. - You pay for no AI plan now and want to start free. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Muse:** Muse uses Muse Spark, Meta's own model. Muse is free with a weekly limit, or $20 or $100 a month. ## At a glance OpenBot is better on 10 of 14 topics, and Muse on 0. OpenBot and Muse compared, as checked on Oct 2, 2026 | Topic | OpenBot | Muse | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Muse Spark, Meta's own model. Meta states no choice of other models. | | Where agents work | Better: On the computer that runs OpenBot. | On a dedicated cloud computer that Meta hosts, the Muse Secure VM. With your permission, Muse for Mac can also use the apps on your Mac. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Muse keeps working after you close the app. You keep no computer on. | | Remote access | Better: Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | You reach Muse through its apps, the web and WhatsApp. Your data stays on your cloud computer at Meta. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Muse uses only Muse Spark, Meta's own model. At the usage limit, you upgrade or wait for the limit to reset. Meta states no backup model. | | Logins and payments | The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | Muse keeps your logins in a store that the agent cannot read. At checkout, it makes a one-time card number, so the agent and the shop never see your card. You approve each purchase. | | Voice | Better: Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Meta showed a Muse that you talk to in real time, but gives no date. Meta states no voice mode in the Muse app today. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | WhatsApp. Meta states no other chat app. | | Teams | Better: Many agents, each with its own job. They give work to each other in shared channels, and coworkers can join one team host. Roles set which agents each member can view, chat with or manage. | One personal agent, with side chats for separate topics. Meta states no team or multi-agent features. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. | Your data and the sign-ins of connected services are on your cloud computer at Meta. By default, Meta can use your Muse conversations to train its AI models; you can turn this off. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Free with a weekly usage limit. Power costs $20 a month and Maximum $100 a month. Needs an account, and is for people aged 18 or more in the US and Canada. | | Where you can use it | Better: In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | Only in the US and Canada, for people 18 and over. Meta says Muse comes to more markets, but gives no countries or dates. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | iPhone, Android, Mac, the web and WhatsApp. Meta's AI glasses are next. A Windows or Linux app: not stated by Meta. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | Not stated by Meta. | Both products give you an AI agent that does the work, not only answers: it keeps a memory, works on long tasks and comes back to you. The difference is who the agent works for, where it runs, and which model does the thinking. Muse is one personal assistant on Meta's cloud. OpenBot is a team of agents on your own computer, with the AI plans you choose. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Muse Muse runs on Muse Spark, Meta's own model, and Meta states no way to use a different one. You pay Meta for use: Muse is free with a weekly limit, and the Power and Maximum plans give more Muse tokens each week. A plan that you already pay for at another AI company does not apply. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Muse Each Muse lives on its own cloud computer, the Muse Secure VM: a Linux computer with a browser and storage that Meta hosts. Your data and the sign-ins of the services you connect are kept there. Muse keeps working after you close the app. With your permission, Muse for Mac can also use the apps on your Mac. ## Phones and remote access ### OpenBot Better The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Muse Muse has apps for iPhone and Android, a web app, a Mac app, and a chat in WhatsApp. Meta says Muse comes to its AI glasses in the coming months. Muse itself and your data stay on your cloud computer at Meta. ## One assistant or a team ### OpenBot Better In OpenBot, each agent has a name, a job and its own workspace. A lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Muse Muse is one personal agent for each person. Side chats keep separate topics apart. Meta describes no way to run several agents, to let agents work together, or to share an agent with coworkers. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds only your profile, team memberships, invitations, sign-in sessions and the settings that let devices find each other. ### Muse Muse keeps your data on your cloud computer at Meta. A separate agent, Sentinel, must approve each action through a connected service and all network traffic, and Muse asks you before sensitive actions such as an email or a purchase. By default, Meta can use your conversations to train its models; you can turn this off in Settings. Meta says it does not give Muse conversations or VM data to its ad systems. A Confidential VM that stops Meta from reading the data in your VM is planned for later in 2026. ## Questions ### Is OpenBot an alternative to Meta's Muse? Yes, if you want your agents on your own computer. Both give you AI agents that do the work and remember it. OpenBot runs a team of agents on your computer with the AI plans you choose. Muse is one personal agent on a cloud computer that Meta hosts, with Meta's own model. ### Why choose OpenBot over Muse? Your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. OpenBot keeps your workspaces, chats and files on your own computer, it runs a team of agents instead of one, and it has desktop apps for macOS, Windows and Linux with no country limit. The app is free, and its source code is on GitHub. Muse is the better fit when you want a personal assistant with no computer to keep on. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can OpenBot run Muse? OpenBot runs Muse Code, Meta's coding agent, as the provider of an agent. Sign in with your Meta account or save a Meta API key, and the agent uses Meta's Muse Spark models next to your other agents. The Muse app, Meta's personal agent on this page, is a separate product, and OpenBot does not run it. ### Which model does Muse use? Muse runs on Muse Spark, Meta's own model. Meta states no way to use a model from a different company. Meta says an open-weights release of Muse Spark is on its roadmap. ### Does Muse run on my computer? Muse lives on a dedicated cloud computer that Meta hosts, the Muse Secure VM. With your permission, Muse for Mac can also use the apps on your Mac. Meta does not state a Windows or Linux app. ### Does Meta train its models on my Muse chats? By default, Meta can use your Muse conversations to train and improve its AI models. You can turn this off in the Muse settings. Meta says it does not give Muse conversations or VM data to its ad systems. OpenBot keeps your chats on your computer, and only the provider you choose gets the requests your agents send. ### Do my chats leave my computer with OpenBot? OpenBot keeps workspaces, conversations and files in a database on the computer that runs it, and OpenBot holds no copy. The provider you choose still gets the requests that your agents send to it. ### Do agents keep working when my laptop is closed? Yes, with both. With OpenBot, run it on a desktop or a server that stays on, or on a hosted OpenBot server, and connect from your laptop or phone. Only that computer must stay on. Muse keeps working on its cloud computer after you close the app. ### Can I use Muse in Europe? Not for now. Meta offers Muse in the US and Canada, to people aged 18 or more, and says it is working to bring Muse to more markets, with no countries or dates. OpenBot has no country limit: it runs on your computer in any country, hosted OpenBot servers run in the EU, and each provider tool follows its own availability. ### How much do OpenBot and Muse cost? OpenBot is free for noncommercial use, and commercial use needs a separate license. Each provider tool uses its own plan or API key. Muse is free with a weekly usage limit. The Power plan costs $20 a month for 500 million Muse tokens a week, and the Maximum plan costs $100 a month for 3 billion. ## Sources Every statement about Muse comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Meta: Introducing Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) - [Meta: Security and safety for AI agents, our approach with Muse](https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse) - [Meta: Introducing Muse Spark 1.3](https://research.meta.ai/blog/introducing-muse-spark-1-3) - [Meta: Everything we announced at Connect 2026](https://www.meta.com/blog/meta-connect-2026-everything-we-announced/) - [Muse for Small Business](https://muse.ai/business) - [Muse subscription plans](https://www.meta.com/help/subscriptions/1021145227643680/) - [Muse side chats](https://www.meta.com/help/artificial-intelligence/1331373868832401/) - [Muse Privacy Policy](https://muse.ai/privacy) - [Muse Terms](https://muse.ai/terms) - [Muse product page](https://ai.meta.com/muse/) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/compare/grok-bot [All comparisons](https://openbot.run/compare) Comparison # OpenBot vs Grok Bot: A Local Grok Bot Alternative OpenBot is a free, local Grok Bot alternative: run AI agents on your computer with the ChatGPT, Claude, Gemini or Grok plan you already pay for. Checked Oct 2, 2026 Sources ## The short answer For most people, OpenBot is the better choice: your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. They run on your own computer, and the app is free. Choose Grok Bot only if you want a cloud computer to run your Bots, so that you keep no computer of your own on. ### Choose OpenBot if Recommended - You already pay for ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code, or you run your own model. - Your files and chats must stay on your own computer. - You want to reach your agents from your phone while they work on your own computer or server. - You want a free app. - You want to read the source code. ### Choose Grok Bot if - You do not want to keep a computer on for your agents. - You already pay for Cursor or SuperGrok. - You want an iPad app. ## Use the plans you already pay for OpenBot sells you no model. Sign in to the AI plans you already have, or connect your own model, and give each agent the one that fits its job. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. **Grok Bot:** Cursor manages model selection. Grok Bot comes with paid Cursor plans and SuperGrok subscriptions. ## At a glance OpenBot is better on 8 of 14 topics, and Grok Bot on 0. OpenBot and Grok Bot compared, as checked on Oct 2, 2026 | Topic | OpenBot | Grok Bot | | --- | --- | --- | | Models and plans | Better: Use the plans you already have: ChatGPT, Claude, Gemini, Grok, Cursor, Cline or Muse Code. Or run OpenCode's free models, Pi, Inception's Mercury models, your own model, or any agent that speaks the Agent Client Protocol. Choose one for each agent, and change it later. | Cursor manages model selection. | | Where agents work | Better: On the computer that runs OpenBot. | On a cloud computer that Cursor hosts, in the United States. With approval, it can also run tasks on your own computer. | | When your computer is off | Agents keep working on the computer or server that runs OpenBot, and routines start them on a schedule. On macOS and Windows, Keep this computer awake stops sleep while it is on power. If you do not want to keep a computer on, use a hosted OpenBot server in the EU, from €30 or $36 a month. | Cloud work continues when you close the app or your laptop. You keep no computer on. | | Remote access | Better: Apps for iPhone and Android, and a web app, connect from anywhere. Chats and files go over an encrypted connection between your devices, with no VPN, and no cloud stores them. Private network mode keeps them on Tailscale or your own Wi-Fi. | The apps connect to the cloud computer that Cursor hosts, where your Bots and their files are. | | Backup models | Better: When a plan reaches its limit, or a provider is overloaded or down, an agent can move to a backup model on another provider. It keeps its chat and workspace, and a note in the chat shows the switch. | Cursor manages model selection, and you cannot choose a model. When weekly usage runs out, Grok Bot stops, unless you turn on on-demand usage. | | Logins and payments | Better: The Vault keeps logins, payment cards and contact info on your computer, encrypted by the operating system. Agents sign in and fill a card for you, but see only the card name and its last four digits, and you approve the first card fill on each site. 1Password and Bitwarden also connect. | Logins stay signed in on the cloud computer, and every Bot of the account shares them. For passwords, 2FA and payment steps, you take over the computer. xAI names no password manager or saved cards. | | Voice | Dictate a message, or talk to an agent in the Voice pane and interrupt it by speaking. Local models turn your speech into text and read replies aloud in 31 languages, so only the text goes to the agent. | Dictate messages, or start a live voice chat on desktop and mobile. A Bot can reply with a voice memo. xAI does not say where it processes speech. | | Chat apps | Agents answer in Slack, Discord and Telegram, and a routine can send its answers there. Connect more than one Slack workspace or Discord server. | Team Bots work in Slack direct messages and channels, and you can tag @bot in a post on X. xAI names no Discord, Telegram or WhatsApp. | | Teams | One computer runs a team host and coworkers join it. Agents give work to each other in shared channels. Roles set which agents each member can view, chat with or manage. | Group chats with two to six Bots, and messages from Bot to Bot. On the Teams plan, every member gets Bots. | | Your data | Better: Workspaces, chats and files stay in a database on your computer. The provider you choose gets the requests you send. | Needs cloud data storage. With Privacy Mode on, customer data is not used for training. | | Price and account | Better: Free for noncommercial use; commercial use needs a license. Your agents use the plans you already pay for. | Included with paid Cursor plans and SuperGrok subscriptions. Needs a Cursor account. | | Where you can use it | In any country: OpenBot has no region lock. The desktop app is in 15 languages, from English, German and Spanish to Japanese, Korean and Turkish, and hosted servers run in the EU. Each AI provider sets the countries for its own plan. | xAI publishes no list of countries for Grok Bot. Its cloud computers run in the United States. The desktop app is in more than 20 languages. | | Apps | macOS, Windows and Linux, mobile apps for iPhone and Android, and a web app. | macOS, Windows and Linux, and mobile apps for iPhone, iPad and Android. | | Source code | Better: On GitHub, under the PolyForm Noncommercial License 1.0.0. | Not stated by xAI. | Both products give you AI agents that stay: each one has a name, a job and its own conversation, and you come back to it as the work grows. In both, agents can pass work to each other, so you do not relay every message yourself. The difference is where that work runs, and how much of it you control. ## Models and the plans you pay for ### OpenBot Better OpenBot runs the provider tools you already use, with the plans you already pay for: Codex with your ChatGPT plan, Claude Code with your Claude plan, Gemini with your Google AI Pro or Ultra plan, Grok CLI with your Grok account or an xAI API key, Cursor CLI with your Cursor plan or a Cursor API key, Cline with your Cline account, and Muse Code with your Meta account or a Meta API key. OpenCode runs free models, or your own model on any OpenAI-compatible server, also one on your computer. Pi runs the providers you sign in to with /login, and Inception's Mercury models run with an Inception API key. Any agent that speaks the Agent Client Protocol, such as Goose, Qwen Code or Oh My Pi, runs as a custom agent. You choose the provider, the model and the reasoning effort for each agent. When a plan reaches its limit or a provider is down, the agent can move to a backup model. When you move an agent to a different provider, it keeps its role, workspace and conversation. ### Grok Bot Grok Bot gives you a managed environment: Cursor manages model selection and the computer. It comes with paid Cursor plans and SuperGrok subscriptions, so you pay for the plan that Grok Bot is part of. There is less to set up, and less of the environment is yours to change. ## Where the work happens ### OpenBot Better OpenBot runs on the computer that hosts it. Agent workspaces, conversations and app data stay on that computer, so it must stay on while its agents work. Run it on a desktop or a server that stays on, and your agents keep working while your laptop is closed. On macOS and Windows, Keep this computer awake stops sleep while the computer is on power, so routines run on time. If you do not want to keep a computer on, use a hosted OpenBot server. It is a Linux server in the EU (Germany, Finland or France), from €30 or $36 a month, and it keeps the workspaces and chats of its agents. Local-first is not offline: an agent that uses a hosted provider still sends its requests to that provider. ### Grok Bot Grok Bot works from a persistent cloud computer that Cursor hosts. One computer belongs to each account, and all of its Bots share it: files, browser sessions and signed-in accounts. Closing the app or your laptop does not stop cloud work. ## Phones and remote access ### OpenBot Better The OpenBot apps for iPhone and Android, and the web app, connect to the computer that runs your agents. From anywhere, you chat with them, follow their progress, approve their actions and send files, and from the phone you can drive their browser. The connection goes directly between your devices when it can, it is encrypted, and no cloud stores your chats or files. You set up no VPN or tunnel. Turn on Private network, and devices connect only over Tailscale or the same Wi-Fi. The phone app locks with Face ID, Touch ID or a PIN. Agents can also answer in Slack, Discord and Telegram. Remote access needs an OpenBot account. ### Grok Bot Grok Bot has apps for iPhone, iPad and Android. Your Bots work on the cloud computer that Cursor hosts, so there is no computer of your own to keep on. Their files, browser sessions and signed-in accounts are on that cloud computer. ## How agents work as a team ### OpenBot In OpenBot, a lead agent can give parts of a task to other agents in a shared channel. You follow their work and step in when a decision needs you. For a team of people, one computer runs the host and the others join it; the chats and files stay on the host. Roles set which agents each member can view, chat with or manage. ### Grok Bot Grok Bot puts two to six Bots in a group chat, and one Bot can send a message to another that it answers later. Because the Bots of an account share one computer, a login or file there is available to every Bot. On the Teams plan, every member has Bots of their own. ## Data and privacy ### OpenBot Better OpenBot keeps workspaces, conversations, attachments and browser data in a SQLite database on the computer that runs it. OpenBot keeps no other copy. On a hosted OpenBot server, the database is on that server in the EU. An account holds only your profile, team memberships, invitations, sign-in sessions and the settings that let devices find each other. ### Grok Bot Grok Bot needs cloud data storage, and its computers run in the United States. With Privacy Mode on, customer data is not used for training. Connector tokens are never stored on the computer, and Cursor keeps a record of Bot actions for 90 days. ## Questions ### Is OpenBot an alternative to Grok Bot? Yes, if you want your agents on your own computer. Both give you persistent AI agents that work as a team. OpenBot runs them on your computer with the provider tools you choose. Grok Bot runs them on a cloud computer that Cursor hosts. ### Why choose OpenBot over Grok Bot? Your agents use the AI plans you already pay for, such as ChatGPT, Claude, Gemini or Grok, or your own model. OpenBot keeps your workspaces, chats and files on your own computer, the app is free, and its source code is on GitHub. You can reach your agents from your phone too. Grok Bot is the better fit only when you do not want to keep any computer on for your agents. ### Can I use my ChatGPT, Claude or Gemini subscription with OpenBot? Yes. OpenBot signs in to each provider tool with your own account: your ChatGPT plan for Codex, your Claude plan for Claude Code, and a Google AI Pro or Ultra plan for Gemini. Grok CLI uses your Grok account or an xAI API key, Cursor CLI your Cursor plan, Cline your Cline account, and Muse Code your Meta account. When a plan reaches its limit, an agent can move to a backup model on another provider. OpenBot adds no charge of its own. ### Can I use my own model with OpenBot? Yes. Through OpenCode, an agent can use any OpenAI-compatible server, also a model that runs on your own computer. You add the address of the server, and an API key if it needs one. ### Can OpenBot use Grok models? Yes. OpenBot can run Grok CLI as the provider of an agent, next to Codex, Claude Code, Gemini, Cursor CLI, Cline, OpenCode, Pi and Muse Code. Grok CLI then sends that agent's requests to xAI. ### Does Grok Bot run on my computer? Grok Bot works from a persistent cloud computer. Through the desktop app, it can also open files and run tasks on your own computer when the settings allow it. By default, it asks before each task. ### Do my chats leave my computer with OpenBot? OpenBot keeps workspaces, conversations and files in a database on the computer that runs it, and OpenBot holds no copy. The provider you choose still gets the requests that your agents send to it. ### Do agents keep working when my laptop is closed? Yes, with both. With OpenBot, run it on a desktop or a server that stays on, or on a hosted OpenBot server, and connect from your laptop or phone. Only that computer must stay on. With Grok Bot, closing the app or your laptop does not stop cloud work. ### Does OpenBot have a mobile app? Yes. The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection. That computer must stay on, and remote access needs an OpenBot account. Your chats and files stay on that computer. ### How much do OpenBot and Grok Bot cost? OpenBot is free for noncommercial use, and commercial use needs a separate license. Each provider tool uses its own plan or API key. Grok Bot is included with paid Cursor plans, the Cursor Teams plan, and SuperGrok subscriptions linked to a Cursor account. ## Sources Every statement about Grok Bot comes from the pages below, checked on Oct 2, 2026. They can change after that date. - [Grok Bot overview](https://docs.x.ai/grok-bot/overview) - [Get started with Grok Bot](https://docs.x.ai/grok-bot/get-started) - [Grok Bot computer and apps](https://docs.x.ai/grok-bot/computer-and-apps) - [Grok Bot chat and collaboration](https://docs.x.ai/grok-bot/chat-and-collaboration) - [Grok Bot for teams and enterprises](https://docs.x.ai/grok-bot/teams-and-enterprises) - [Grok Bot settings and languages](https://docs.x.ai/grok-bot/settings-and-notifications) - [Grok Bot security](https://docs.x.ai/grok-bot/security) - [Grok Bot plans and billing](https://cursor.com/help/grok-bot/plans) - [Grok Bot on mobile](https://docs.x.ai/grok-bot/mobile) - [Grok Bot Team Bots](https://docs.x.ai/grok-bot/team-bots) - [Grok Bot tag on X](https://docs.x.ai/grok-bot/tag-on-x) - [OpenBot privacy notes](https://github.com/nightly-labs/openbot/blob/main/PRIVACY.md) - [OpenBot source code](https://github.com/nightly-labs/openbot) --- Source: https://openbot.run/providers/muse-code [All providers](https://openbot.run/providers) Provider # Muse Code GUI: Run Meta's Coding Agent in OpenBot OpenBot runs Muse Code, Meta's coding agent, as one agent of a team, on macOS, Windows or Linux, with your Meta account or a Meta API key. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 24, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Muse Code Choose Muse, and OpenBot downloads and pins its own copy of Muse Code. Sign in with your Meta account in the browser, or save a Meta API key. 3. ### Give your agent a job Create an agent on Muse Code, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Meta's models A Muse Code agent uses the Meta models that Muse Code lists, such as Muse Spark 1.3. - ### Steer it while it works Send a message while a Muse Code agent works, and the message changes the course of the current turn. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Meta, and where they run. OpenBot runs the same agent, so you can use both. Apps from Meta, as checked on Oct 24, 2026 | App | Runs on | What it is | | --- | --- | --- | | Muse Code in the terminal | macOS, Linux, Windows | The terminal agent. `muse exec` runs it with no interface, and `muse serve` runs it as a server. | | Muse Code SDK | TypeScript. In developer preview | The engine behind the CLI, for your own program. | ## Every plan an agent can use Muse Code is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Muse Code Your Meta account, or a Meta API key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [OpenBot vs Muse: A Local Alternative to Meta's Muse](https://openbot.run/compare/muse) ## Questions ### Is Muse Code the same as the Muse app? No. Muse Code is Meta's coding agent for the terminal. The Muse app is Meta's personal agent on a cloud computer. OpenBot runs Muse Code, and OpenBot vs Muse compares OpenBot with the app. ### Is there a GUI for Muse Code? Meta ships Muse Code for the terminal and as an SDK, with no desktop app. OpenBot is a desktop app that runs Muse Code as one agent of a team, with agents on other providers and an app on your phone. ### How much does Muse Code cost? Meta's plans cost $5, $15, $50 and $200 a month, and each one gives more use. By default, a plan is in Contributor mode: you get more use, and Meta can train on your content. You can also pay for each token with a Meta API key. Prices can be different in your region. ## Sources Every statement about Muse Code comes from the pages below, checked on Oct 24, 2026. They can change after that date. - [Meta: Muse Code](https://dev.meta.ai/products/muse-code/) - [Muse Code documentation](https://dev.meta.ai/docs/muse-code/) - [Muse Code sign-in](https://dev.meta.ai/docs/muse-code/auth) - [Muse Code subscriptions](https://dev.meta.ai/docs/muse-code/subscriptions) - [Meta Model API prices and rate limits](https://dev.meta.ai/docs/pricing-rate-limits) - [Meta: Muse Code plans and features](https://dev.meta.ai/resources/blog/muse-code-new-plans-and-features) --- Source: https://openbot.run/providers/pi [All providers](https://openbot.run/providers) Provider # Pi Coding Agent GUI: Run Pi in OpenBot OpenBot runs Pi as one agent of a team, on macOS, Windows or Linux, with the providers that you sign in to in Pi. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 24, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Pi Choose Pi, and OpenBot downloads and pins its own copy of Pi. To sign in, run `/login` in Pi on the computer that runs OpenBot, connect a subscription or an API key, and close Pi. OpenBot shows Pi as connected soon after. 3. ### Give your agent a job Create an agent on Pi, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### The models you sign in to A Pi agent uses the models that Pi lists for the providers that you connected with /login. - ### Steer it while it works Send a message while a Pi agent works, and the message changes the course of the current turn. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Earendil, and where they run. OpenBot runs the same agent, so you can use both. Apps from Earendil, as checked on Oct 24, 2026 | App | Runs on | What it is | | --- | --- | --- | | Pi in the terminal | macOS, Linux, Windows. Native Windows uses Git Bash | The terminal agent. | | Pi SDK | TypeScript | Pi in your own program. | ## Every plan an agent can use Pi is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a GUI for Pi? Pi has no official desktop or web app: it works in the terminal. OpenBot is a desktop app that runs Pi as one agent of a team, with agents on other providers and an app on your phone. ### How do I sign in to Pi in OpenBot? Run /login in Pi on the computer that runs OpenBot, and connect a subscription or an API key. Then close Pi, and OpenBot shows Pi as connected. ### Is Pi free? Pi is open source under the MIT license, and OpenBot downloads it for you. You pay the providers that you sign in to. ### Does Pi run Inception's Mercury models? Yes. In OpenBot, Inception runs in the copy of Pi that OpenBot downloads, with an Inception API key that OpenBot keeps encrypted. Pi's own settings do not change. ## Sources Every statement about Pi comes from the pages below, checked on Oct 24, 2026. They can change after that date. - [Pi](https://pi.dev/) - [Pi documentation](https://pi.dev/docs/latest) - [Pi on GitHub](https://github.com/earendil-works/pi) - [Pi on Windows](https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/windows.md) - [Earendil: Pi 1.0](https://earendil.com/posts/pi-1-0/) --- Source: https://openbot.run/providers/inception [All providers](https://openbot.run/providers) Provider # Mercury Agent App: Run Inception's Mercury Models in OpenBot OpenBot runs Inception's Mercury models as one agent of a team, on macOS, Windows or Linux, with an Inception API key. You get shared channels with agents on other providers. Checked Oct 24, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Inception Download Pi in OpenBot first: Mercury runs in OpenBot's copy of Pi. Then select Connect on the Inception row and paste an Inception API key. OpenBot keeps the key encrypted and does not change Pi's own settings. 3. ### Give your agent a job Create an agent on Inception, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Three Mercury models Choose Mercury 3, with a context of 1 million tokens, or Mercury 2.5 or Mercury 2. - ### Your key stays in OpenBot OpenBot keeps the Inception API key in its encrypted store, not in Pi's settings files. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Inception, and where they run. OpenBot runs the same agent, so you can use both. Apps from Inception, as checked on Oct 24, 2026 | App | Runs on | What it is | | --- | --- | --- | | Mercury Chat | Your browser | A chat with the Mercury models. | | Inception API Platform | Your browser, and any OpenAI-compatible client | API keys and use of the Mercury models. | ## Every plan an agent can use Inception is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Inception Mercury models, with an Inception API key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### What is Mercury? Mercury is Inception's family of diffusion language models. They make tokens in parallel, at more than 1,000 tokens a second, Inception says. Mercury 3 came out on 19 October 2026, with a context of 1 million tokens. ### How much does Mercury cost? For each million tokens, Mercury 3 costs $0.10 for input and $0.35 for output, at 50% off for its launch. Mercury 2.5 costs $0.20 and $0.75, and Mercury 2 costs $0.25 and $0.75. New accounts get 100 million free tokens. ### Why does Inception need Pi? OpenBot runs Mercury in its own copy of Pi. Download Pi once in OpenBot, and Inception agents then use it with your Inception API key. ### Can I use Inception agents from my phone or share them with a team? Yes. An Inception agent runs on the computer that has your Inception API key, and you use it from your phone or share it with your team like any other agent. On a server, an owner or admin saves the key in the server's provider settings. ## Sources Every statement about Inception comes from the pages below, checked on Oct 24, 2026. They can change after that date. - [Inception](https://inceptionlabs.ai/) - [Inception: models and prices](https://docs.inceptionlabs.ai/get-started/models) - [Inception: get started](https://docs.inceptionlabs.ai/get-started/get-started) - [Inception: introducing Mercury 3](https://www.inceptionlabs.ai/blog/introducing-mercury-3) --- Source: https://openbot.run/providers/custom-agents [All providers](https://openbot.run/providers) Provider # ACP Agent GUI: Run Goose, Qwen Code and Oh My Pi in OpenBot OpenBot runs any agent that speaks the Agent Client Protocol as one agent of a team, such as Goose, Qwen Code or Oh My Pi. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 24, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Goose, Qwen Code or Oh My Pi Add a custom agent and type the command that starts it, or pick an agent that OpenBot finds on your computer or in the ACP registry. Check agent starts the command once to make sure that the agent answers. Each agent signs in with its own command. 3. ### Give your agent a job Create an agent on Goose, Qwen Code or Oh My Pi, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Presets for known agents Presets fill in the start command for Goose, Qwen Code and Oh My Pi. - ### The ACP registry Search, install, update and remove agents from the ACP registry on the computer that runs OpenBot. - ### Encrypted settings OpenBot keeps the environment variables of each agent, such as an API key, encrypted. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from their makers, and where they run. OpenBot runs the same agent, so you can use both. Apps from their makers, as checked on Oct 24, 2026 | App | Runs on | What it is | | --- | --- | --- | | Goose | macOS, Linux, Windows | An open-source agent with a desktop app and a CLI. Apache 2.0. | | Qwen Code | Terminal | Qwen's coding agent. Its free tier ended on 2026-04-15. | | Oh My Pi | Terminal | A fork of Pi. MIT license. | ## Every plan an agent can use Goose, Qwen Code or Oh My Pi is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Your own model Any OpenAI-compatible server, also one on your computer. ## Questions ### What is the Agent Client Protocol? The Agent Client Protocol (ACP) is an open protocol between apps and coding agents. A local agent talks JSON-RPC over its standard input and output. Zed and JetBrains govern it. ### Which agents work in OpenBot? Any agent that speaks ACP over standard input and output. OpenBot has presets for Goose, Qwen Code and Oh My Pi, and finds more in the ACP registry. ### Can I send a message while a custom agent works? Yes. The message waits until the current turn ends, and then the agent gets it. A custom agent works on one turn at a time. ### Can I use GitHub Copilot? Yes. GitHub Copilot is a built-in provider in OpenBot, not a custom agent. Connect it in Settings > Providers. ## Sources Every statement about Goose, Qwen Code or Oh My Pi comes from the pages below, checked on Oct 24, 2026. They can change after that date. - [Agent Client Protocol: introduction](https://agentclientprotocol.com/get-started/introduction) - [Agent Client Protocol: agents](https://agentclientprotocol.com/get-started/agents) - [Agent Client Protocol: registry](https://agentclientprotocol.com/get-started/registry) - [Agent Client Protocol: governance](https://agentclientprotocol.com/community/governance) - [Goose](https://goose-docs.ai/) - [Qwen Code sign-in](https://qwenlm.github.io/qwen-code-docs/en/users/configuration/auth/) - [Oh My Pi on GitHub](https://github.com/can1357/oh-my-pi) --- Source: https://openbot.run/providers/claude-code [All providers](https://openbot.run/providers) Provider # Claude Code GUI: Run Claude Code in a Desktop App OpenBot runs Claude Code as one agent of a team, on macOS, Windows or Linux, with the Claude plan you already pay for. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Claude Code Choose Claude, and OpenBot downloads and pins its own copy of Claude Code, or uses the one you installed. Then sign in with your Claude account. If `claude` works in your terminal, it works in OpenBot. 3. ### Give your agent a job Create an agent on Claude Code, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### A full desktop app on Linux Anthropic's desktop app is in beta on Linux, for Ubuntu and Debian only. OpenBot runs Claude Code on x64 and arm64 Linux, as an AppImage. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Anthropic, and where they run. OpenBot runs the same agent, so you can use both. Apps from Anthropic, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Claude desktop app, Code tab | macOS, Windows. Linux in beta, Ubuntu 22.04+ and Debian 12+ | Claude Code with a graphical interface. Needs a paid Claude plan. | | Claude Code command line | macOS 13+, Windows 10+, Ubuntu, Debian, Alpine | The terminal agent. No WSL is necessary on Windows. | | IDE extensions | VS Code and its forks, JetBrains IDEs | Claude Code in the editor you already use. | | Claude Code on the web | Browser, and the Claude apps for iPhone and Android | Cloud sessions on Pro, Max, Team and Enterprise plans. | ## Every plan an agent can use Claude Code is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Claude Your Claude plan, through Claude Code. - ChatGPT Your ChatGPT plan, through Codex. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Codex vs Claude Code: Which Coding Agent to Use](https://openbot.run/compare/codex-vs-claude-code) - [Cursor vs Claude Code: Editor or Terminal Agent](https://openbot.run/compare/cursor-vs-claude-code) - [Claude Code vs Antigravity: Claude or Gemini Agent](https://openbot.run/compare/claude-code-vs-antigravity) - [OpenBot vs Claude Cowork: A Local Alternative](https://openbot.run/compare/claude-cowork) ## Questions ### Is there a GUI for Claude Code? Yes. Anthropic's Claude desktop app has a Code tab for macOS and Windows, and Linux in beta. OpenBot is a different desktop app: it runs Claude Code as one agent of a team, with agents on other providers and an app on your phone. ### Does OpenBot need an Anthropic API key for Claude Code? No. OpenBot uses the sign-in of Claude Code, so your Claude Pro or Max plan works. OpenBot does not copy your credentials. The free Claude plan does not include Claude Code. ### Can I run Claude Code on Linux with a graphical app? Yes. Anthropic's desktop app is in beta on Linux, for Ubuntu 22.04 or later and Debian 12 or later. OpenBot runs Claude Code on x64 and arm64 Linux as an AppImage, on any distribution that runs an AppImage. ### Can Claude Code and Codex work together in OpenBot? Yes. Each agent has its own provider. A Claude Code agent and a Codex agent can work in the same channel and give tasks to each other. ## Sources Every statement about Claude Code comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Claude Code: desktop app](https://code.claude.com/docs/en/desktop) - [Claude Code: desktop app on Linux](https://code.claude.com/docs/en/desktop-linux) - [Claude Code: set up](https://code.claude.com/docs/en/setup) - [Claude Code: overview](https://code.claude.com/docs/en/overview) - [Claude Code on the web](https://code.claude.com/docs/en/claude-code-on-the-web) - [Claude pricing](https://claude.com/pricing) --- Source: https://openbot.run/providers/codex [All providers](https://openbot.run/providers) Provider # Codex Desktop App for macOS, Windows and Linux OpenBot runs OpenAI Codex as one agent of a team, on macOS, Windows or Linux, with the ChatGPT plan you already pay for. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Codex Choose ChatGPT, and OpenBot downloads and pins its own copy of Codex, or uses the one you installed. Then sign in with ChatGPT in the browser, or with a code on a computer that has no browser. 3. ### Give your agent a job Create an agent on Codex, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### The same app on every desktop OpenBot runs Codex on macOS on Apple silicon and Intel, on Windows 10 or later, and on x64 or arm64 Linux, on any distribution that runs an AppImage. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from OpenAI, and where they run. OpenBot runs the same agent, so you can use both. Apps from OpenAI, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | ChatGPT desktop app, Codex mode | macOS on Apple silicon, Windows. Linux in preview | Codex with a graphical interface. Native on Windows, with no WSL. | | Codex command line | macOS, Linux, Windows | The terminal agent. Windows 11 is recommended; Windows 10 is best effort. | | IDE extension | VS Code, Cursor, Windsurf, JetBrains IDEs | Codex in the editor you already use. | | Codex in ChatGPT | Web, and the ChatGPT apps for phones | Cloud tasks that run on OpenAI's servers. | ## Every plan an agent can use Codex is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Codex vs Claude Code: Which Coding Agent to Use](https://openbot.run/compare/codex-vs-claude-code) - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a Codex desktop app for Windows and Linux? Yes. OpenAI's ChatGPT desktop app has a Codex mode on Windows, and on Linux in preview. OpenBot is a different desktop app: it runs Codex on macOS, Windows and Linux, as one agent of a team. ### Does Codex in OpenBot need an OpenAI API key? No. You sign in with ChatGPT, so your ChatGPT plan pays for the work. OpenAI includes Codex in the Free, Go, Plus, Pro, Business, Edu and Enterprise plans, with limits for each plan. ### Does Codex need WSL on Windows? No. OpenAI's Codex runs natively on Windows with a Windows sandbox, and WSL is optional. OpenBot runs Codex on Windows 10 or later, on x64. ### Can Codex and Claude Code work together in OpenBot? Yes. Each agent has its own provider. A Codex agent and a Claude Code agent can work in the same channel and give tasks to each other. ## Sources Every statement about Codex comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Codex: the desktop app](https://learn.chatgpt.com/docs/app) - [Codex: the Windows app](https://learn.chatgpt.com/docs/windows/windows-app) - [Codex: the Windows sandbox](https://learn.chatgpt.com/docs/windows/windows-sandbox) - [Codex: the Linux app](https://learn.chatgpt.com/docs/linux/linux-app) - [Codex: command line](https://learn.chatgpt.com/docs/codex/cli) - [Codex: IDE extension](https://learn.chatgpt.com/docs/codex/ide) - [Codex pricing](https://learn.chatgpt.com/docs/pricing) --- Source: https://openbot.run/providers/gemini [All providers](https://openbot.run/providers) Provider # Gemini Agent Desktop App: Run Gemini in OpenBot OpenBot runs Gemini as one agent of a team, on macOS, Windows or Linux, through Google's Antigravity agent and your Google AI Pro or Ultra plan. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Gemini Choose Gemini, and OpenBot downloads and pins its own copy of Google's Antigravity agent server. Then sign in with the Google account that has your Google AI Pro or Ultra plan. 3. ### Give your agent a job Create an agent on Gemini, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Gemini next to other providers Antigravity runs Google's agents. In OpenBot, a Gemini agent works in the same channel as Claude Code, Codex or Grok agents. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Google, and where they run. OpenBot runs the same agent, so you can use both. Apps from Google, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Antigravity 2.0 | macOS 12+, Windows 10+, Linux, on x64 and arm64 | A desktop app that manages many local agents at once. | | Antigravity IDE | macOS 12+, Windows 10+, Linux, on x64 and arm64 | An editor with agents built in. | | Antigravity command line | macOS, Windows, Linux | The terminal agent, `agy`. | | Gemini desktop app | macOS 15+ on Apple silicon, Windows 10+ | Chat with Gemini. Not a coding agent. | ## Every plan an agent can use Gemini is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Gemini Your Google AI Pro or Ultra plan. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Claude Code vs Antigravity: Claude or Gemini Agent](https://openbot.run/compare/claude-code-vs-antigravity) - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a Gemini agent desktop app? Yes. Google's Antigravity 2.0 is a desktop app for coding agents on macOS, Windows and Linux. OpenBot is a different desktop app: it runs Gemini as one agent of a team, with agents on other providers and an app on your phone. ### Which Google plan does Gemini in OpenBot need? Google AI Pro or Ultra. Google gives the Pro plan entry rate limits for agents in Antigravity, which refresh every five hours up to a weekly limit. Ultra has five or twenty times the usage of Pro. ### Does OpenBot use Gemini CLI? No. On 18 June 2026, Gemini CLI stopped serving Google AI Pro, Ultra and free accounts. OpenBot runs Gemini through Google's Antigravity agent server, which those plans use now. ### Can Gemini and Claude Code work together in OpenBot? Yes. Each agent has its own provider. A Gemini agent and a Claude Code agent can work in the same channel and give tasks to each other. ## Sources Every statement about Gemini comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Google Antigravity](https://antigravity.google/) - [Antigravity: download](https://antigravity.google/download) - [Antigravity: get started with the command line](https://antigravity.google/docs/getting-started?tab=cli) - [Antigravity: plans](https://antigravity.google/docs/plans) - [Gemini CLI: the move to Antigravity](https://github.com/google-gemini/gemini-cli/discussions/27274) - [Google AI plans](https://gemini.google/us/subscriptions/?hl=en) - [Gemini for desktop](https://gemini.google/desktop/) --- Source: https://openbot.run/providers/grok [All providers](https://openbot.run/providers) Provider # Grok CLI GUI: Run the Grok Coding Agent in OpenBot OpenBot gives xAI's Grok coding agent a desktop app on macOS, Windows or Linux, with your Grok account or an xAI API key. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Grok Choose Grok, and OpenBot downloads and pins its own copy of the Grok CLI, or uses the one you installed. Then sign in with your Grok account, or set `XAI_API_KEY`. 3. ### Give your agent a job Create an agent on Grok, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### A window for a terminal agent Grok Build runs in the terminal, or in an editor that speaks the Agent Client Protocol. OpenBot gives it a desktop app with threads, files and a model picker. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from xAI, and where they run. OpenBot runs the same agent, so you can use both. Apps from xAI, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Grok Build | macOS, Linux, Windows | The coding agent, in the terminal or headless. It also speaks the Agent Client Protocol. | | Grok Bot | macOS, Windows, Linux | A desktop agent for general tasks, not coding. Needs a paid Cursor plan or a SuperGrok plan linked to Cursor. | ## Every plan an agent can use Grok is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Grok Your Grok account or an xAI API key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [OpenBot vs Grok Bot: A Local Grok Bot Alternative](https://openbot.run/compare/grok-bot) - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a GUI for the Grok CLI? xAI documents no graphical app for Grok Build: it runs in the terminal, in scripts, or in an editor through the Agent Client Protocol. OpenBot runs it in a desktop app on macOS, Windows and Linux. ### Which plan does Grok in OpenBot need? xAI says you can try Grok Build for free. SuperGrok, at $30 a month, has higher limits, and SuperGrok Plus, at $100, has much more usage of Grok Build. An xAI API key also works, at API prices. ### Is Grok Bot the same as Grok Build? No. Grok Bot is xAI's desktop agent for general tasks on your computer. Grok Build is xAI's coding agent, and it is the one OpenBot runs. ### Can Grok and Claude Code work together in OpenBot? Yes. Each agent has its own provider. A Grok agent and a Claude Code agent can work in the same channel and give tasks to each other. ## Sources Every statement about Grok comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Grok Build: overview](https://docs.x.ai/build/overview) - [Grok Build](https://x.ai/build) - [xAI: Grok Build CLI](https://x.ai/news/grok-build-cli) - [xAI pricing](https://x.ai/pricing) - [Grok Bot: get started](https://docs.x.ai/grok-bot/get-started) --- Source: https://openbot.run/providers/cursor [All providers](https://openbot.run/providers) Provider # Cursor CLI GUI: Run the Cursor Agent in OpenBot OpenBot runs the Cursor agent outside the Cursor editor, on macOS, Windows or Linux, with your Cursor plan or a Cursor API key. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Cursor Choose Cursor, and OpenBot downloads and pins its own copy of the Cursor CLI, or uses the one you installed. Then sign in with Cursor in the browser, or set `CURSOR_API_KEY`. 3. ### Give your agent a job Create an agent on Cursor, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### The Cursor agent without the editor The Cursor CLI runs in the terminal. OpenBot gives it a desktop app with threads and files, so you can use Cursor's models without the editor open. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Cursor, and where they run. OpenBot runs the same agent, so you can use both. Apps from Cursor, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Cursor editor | macOS, Windows, Linux, on x64 and arm64 | An editor built from VS Code, with agents built in. | | Cursor CLI | macOS, Linux, Windows, WSL | The agent in the terminal, as `agent`. It also speaks the Agent Client Protocol. | ## Every plan an agent can use Cursor is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Cursor Your Cursor plan, or a Cursor API key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Cursor vs Claude Code: Editor or Terminal Agent](https://openbot.run/compare/cursor-vs-claude-code) - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a GUI for the Cursor CLI? The Cursor editor is Cursor's own graphical app. The Cursor CLI also runs in editors that speak the Agent Client Protocol, such as Zed and JetBrains IDEs. OpenBot is a desktop app that runs it as one agent of a team. ### Which Cursor plan does OpenBot need? Your Cursor plan, or a Cursor API key from the Cursor dashboard. Hobby is free; Pro is $20 a month, Pro Plus $60 and Ultra $200. Use beyond the plan is billed at API prices. ### Do I need the Cursor editor to use Cursor in OpenBot? No. OpenBot runs the Cursor CLI, which is a separate program. The editor does not have to be installed or open. ### Can Cursor and Claude Code work together in OpenBot? Yes. Each agent has its own provider. A Cursor agent and a Claude Code agent can work in the same channel and give tasks to each other. ## Sources Every statement about Cursor comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Cursor: download](https://cursor.com/download) - [Cursor CLI: install](https://cursor.com/docs/cli/installation) - [Cursor CLI: Agent Client Protocol](https://cursor.com/docs/cli/acp) - [Cursor CLI: authentication](https://cursor.com/docs/cli/reference/authentication) - [Cursor: plans and usage](https://cursor.com/docs/account/pricing) - [Cursor pricing](https://cursor.com/pricing) --- Source: https://openbot.run/providers/opencode [All providers](https://openbot.run/providers) Provider # OpenCode GUI: Run OpenCode and Its Free Models in OpenBot OpenBot runs OpenCode as one agent of a team, on macOS, Windows or Linux, with its free models and no account. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect OpenCode Choose OpenCode, and OpenBot downloads and pins its own copy of OpenCode, or uses the one you installed. The free models need no sign-in. For more models, add an OpenCode Go key. 3. ### Give your agent a job Create an agent on OpenCode, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Free to start OpenCode's free models need no account and no key, so a new agent works as soon as OpenCode is downloaded. - ### Your own models OpenBot runs OpenCode on any OpenAI-compatible server too, such as Ollama or LM Studio on your computer. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from OpenCode, and where they run. OpenBot runs the same agent, so you can use both. Apps from OpenCode, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | OpenCode in the terminal | macOS, Linux, Windows. WSL is recommended on Windows | The terminal agent. It also speaks the Agent Client Protocol. | | OpenCode desktop app | macOS, Windows x64, Linux. In beta | OpenCode in a window. | | OpenCode web | Your browser | `opencode web` starts a local server and opens it in the browser. | | IDE extensions | VS Code, Cursor, Zed, Windsurf, VSCodium | OpenCode in the editor you already use. | ## Every plan an agent can use OpenCode is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - OpenCode Free models, or an OpenCode Go key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a GUI for OpenCode? Yes. OpenCode has a desktop app in beta and a web interface that runs on your computer. OpenBot is a different desktop app: it runs OpenCode as one agent of a team, with agents on other providers and an app on your phone. ### Is OpenCode in OpenBot free? Yes, with OpenCode's free models: they need no account. OpenCode Go, at $10 a month, and Go Plus, at $40, give reliable use of open coding models. Both are optional. ### Does OpenCode need WSL on Windows? OpenCode recommends WSL for its own terminal app on Windows. OpenBot downloads the native Windows x64 build of OpenCode and runs it for you, with no WSL. ### Can OpenCode run a model on my computer? Yes. In OpenBot, OpenCode runs any OpenAI-compatible server as a custom provider, such as Ollama or LM Studio on your computer. ## Sources Every statement about OpenCode comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [OpenCode: introduction](https://opencode.ai/docs/) - [OpenCode: download](https://opencode.ai/download) - [OpenCode: web](https://opencode.ai/docs/web/) - [OpenCode: Agent Client Protocol](https://opencode.ai/docs/acp/) - [OpenCode Zen](https://opencode.ai/docs/zen/) - [OpenCode Go](https://opencode.ai/docs/go/) - [OpenCode on GitHub](https://github.com/anomalyco/opencode) --- Source: https://openbot.run/providers/cline [All providers](https://openbot.run/providers) Provider # Cline Desktop App: Run Cline Without an Editor in OpenBot OpenBot runs the Cline agent without an editor, on macOS, Windows or Linux, with your Cline account or a Cline API key. You get shared channels with agents on other providers, and an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Cline Choose Cline, and OpenBot downloads and pins its own copy of the Cline CLI, or uses the one you installed, version 3.0.68 or later. Then sign in with Cline in the browser, or set `CLINE_API_KEY`. 3. ### Give your agent a job Create an agent on Cline, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### A window for the Cline CLI The Cline CLI runs in the terminal. OpenBot gives it a desktop app with threads and files, and no editor is necessary. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Cline, and where they run. OpenBot runs the same agent, so you can use both. Apps from Cline, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Cline extension | VS Code, Cursor, Windsurf, VSCodium, JetBrains IDEs | The agent in the editor you already use. | | Cline for Desktop | macOS, Windows, and Linux on the download page. In beta | The Cline agent as a desktop app, with no editor. | | Cline CLI | Any computer with Node.js 20 or later | The agent in the terminal. It also speaks the Agent Client Protocol. | ## Every plan an agent can use Cline is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Cline Your Cline account, or a Cline API key. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. - Your own model Any OpenAI-compatible server, also one on your computer. ## Comparisons - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Is there a Cline desktop app? Yes. Cline for Desktop is in beta. Its download page lists macOS, Windows and Linux, but the Cline documentation names only macOS and Windows. OpenBot is a different desktop app: it runs the Cline CLI as one agent of a team. ### What does Cline cost in OpenBot? Cline is open source under the Apache 2.0 license, and has no subscription: you pay for AI use. ClinePass, at $9.99 a month, is optional. OpenBot uses your Cline account or a Cline API key. ### Do I need VS Code to use Cline in OpenBot? No. OpenBot runs the Cline CLI, which needs no editor. ### Can Cline and Claude Code work together in OpenBot? Yes. Each agent has its own provider. A Cline agent and a Claude Code agent can work in the same channel and give tasks to each other. ## Sources Every statement about Cline comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Cline for Desktop](https://cline.bot/desktop) - [Cline: install](https://docs.cline.bot/getting-started/installing-cline) - [Cline CLI: overview](https://docs.cline.bot/cline-cli/overview) - [Cline: Agent Client Protocol](https://docs.cline.bot/usage/acp) - [ClinePass](https://docs.cline.bot/getting-started/clinepass) - [Cline pricing](https://cline.bot/pricing) - [Cline on GitHub](https://github.com/cline/cline) --- Source: https://openbot.run/providers/local-models [All providers](https://openbot.run/providers) Provider # Ollama and LM Studio Agent App: Run Local Models in OpenBot OpenBot runs a coding agent on a model from Ollama or LM Studio, on macOS, Windows or Linux. The model, your files and your chats stay on your computer, and your agents work in a team with an app on your phone. Checked Oct 3, 2026 Sources ## How to set it up 1. ### Download OpenBot For macOS 13 or later on Apple silicon or Intel, Windows 10 or later on x64, or x64 or arm64 Linux. [macOS](https://openbot.run/download/macos)[Windows](https://openbot.run/download/windows)[Linux](https://openbot.run/download/linux) 2. ### Connect Ollama and LM Studio Start Ollama or LM Studio. OpenBot looks at their usual addresses, 127.0.0.1:11434 and 127.0.0.1:1234, and shows the server under Found on this computer, in onboarding and in Settings. Press Add, and OpenCode runs your agents on its models. 3. ### Give your agent a job Create an agent on Ollama and LM Studio, choose its model, and give it a task. OpenBot makes a workspace folder for it. When the work grows, add agents on other providers to the same channel. ## What OpenBot adds - ### Model requests stay on your computer With a local model, the requests go to a server on your computer, not to a provider. Any other OpenAI-compatible server works too. - ### Local and hosted models in one team Give the simple jobs to a local model, and the hard ones to Claude Code or Codex, in the same channel. - ### A team, not one chat Each agent has its own job and workspace. A lead agent gives parts of a task to other agents in a shared channel, and you step in when a decision needs you. - ### Your agents on your phone The OpenBot apps for iPhone and Android connect to the computer that runs your agents, over an encrypted connection, with no VPN. Remote access needs an OpenBot account. - ### Change the provider, keep the work Move an agent to a different provider and it keeps its role, workspace and conversation. Agents on different providers work in the same channel. When a plan reaches its limit, an agent can move to a backup model. - ### Your data on your computer Workspaces, chats and files stay on the computer that runs OpenBot. The provider you choose gets the requests you send. - ### Use your AI plans OpenBot is free for noncommercial use. Your agents use the plans you already pay for. ## Official apps The apps from Ollama and LM Studio, and where they run. OpenBot runs the same agent, so you can use both. Apps from Ollama and LM Studio, as checked on Oct 3, 2026 | App | Runs on | What it is | | --- | --- | --- | | Ollama | macOS 14+, Windows 10 22H2+, Linux | A model server with a chat app on macOS and Windows. Its OpenAI-compatible API supports tools. | | LM Studio | macOS 14+ on Apple silicon, Windows x64 and Arm, Linux x64 and arm64 | A desktop app to download and chat with models, with a local server. Free at home and at work. | ## Every plan an agent can use Ollama and LM Studio is one of them. Choose a provider for each agent, and change it later: the agent keeps its role, workspace and conversation. - Your own model Any OpenAI-compatible server, also one on your computer. - ChatGPT Your ChatGPT plan, through Codex. - Claude Your Claude plan, through Claude Code. - Gemini Your Google AI Pro or Ultra plan. - Grok Your Grok account or an xAI API key. - Cursor Your Cursor plan, or a Cursor API key. - OpenCode Free models, or an OpenCode Go key. - Cline Your Cline account, or a Cline API key. - Muse Code Your Meta account, or a Meta API key. - Pi Included with OpenBot. Sign in to a provider with /login in Pi. - Inception Mercury models, with an Inception API key. - Custom agent Any ACP agent, such as Goose, Qwen Code or Oh My Pi. ## Comparisons - [Best AI Agent Apps in 2026: 13 Apps Compared](https://openbot.run/compare/best-ai-agent-apps) ## Questions ### Can I use Ollama or LM Studio with a coding agent? Yes. In OpenBot, OpenCode runs the agent on the OpenAI-compatible server of Ollama or LM Studio. Choose a model that supports tool calls: smaller models and models not trained for tools can send tool calls in the wrong format. ### How much context does a local model need? More than the default. Ollama says that agents and coding tools need at least 64,000 tokens of context, and its default can be 4,000 tokens on a smaller graphics card. Set it higher in Ollama before you start. ### Does a local model need an account? No. Ollama and LM Studio need no account for local models. ### Can I use a model server on another computer? Yes. Add the address of any OpenAI-compatible server in OpenBot's settings, with a key if the server needs one. ## Sources Every statement about Ollama and LM Studio comes from the pages below, checked on Oct 3, 2026. They can change after that date. - [Ollama: download](https://ollama.com/download) - [Ollama: macOS](https://docs.ollama.com/macos) - [Ollama: Windows](https://docs.ollama.com/windows) - [Ollama: OpenAI compatibility](https://docs.ollama.com/api/openai-compatibility) - [Ollama: context length](https://docs.ollama.com/context-length) - [Ollama: the new app](https://ollama.com/blog/new-app) - [LM Studio: system requirements](https://lmstudio.ai/docs/app/system-requirements) - [LM Studio: OpenAI compatibility](https://lmstudio.ai/docs/developer/openai-compat) - [LM Studio: tool use](https://lmstudio.ai/docs/developer/openai-compat/tools) - [LM Studio: free for work](https://lmstudio.ai/blog/free-for-work)