LATEST
AI Tools

OpenBot 1.0: Free AI Coworkers, $29 for Teams, One License Catch

Updated November 16, 2026
17 min read
Editor's Note: This article was human-reviewed for accuracy
A human hand reaching toward a white robotic hand against a pale blue background, symbolizing people working alongside AI coworkers

Quick Answer & Key Takeaways

  • • CopilotKit released OpenBot 1.0 on November 9, 2026, a free workspace where AI coworkers get their own computer, browser, and tools.
  • • Personal use is free, while OpenBot for Business costs up to $29 per active user per month, plus a 5% charge on model spending above $5,000 a month.
  • • The application template is MIT-licensed, but the CopilotKit Intelligence layer it runs on is licensed separately.

CopilotKit released OpenBot 1.0 on November 9, 2026, and the pitch is simple: an AI coworker that works on a computer of its own, on your machine, under rules you set. Each Bot gets its own browser, files, and tools. It asks before it deletes, sends, shares, or posts anything, and every action lands in an audit trail. The desktop app is free for personal use, the code template carries an MIT license, and the project passed 12,000 GitHub stars within a week of launch. That combination is rare enough to take seriously. The catch is in the fine print, and it is the first thing to understand before you install it.

OpenBot comes from CopilotKit, the team behind the AG-UI agent protocol. Version 1.0.0 was tagged on GitHub on November 9, 2026, with CopilotKit's announcement posts following on November 10. Three point releases followed in five days, taking it to v1.0.3 on November 14. The repository, CopilotKit/OpenBot, showed 12,399 stars and 1,532 forks when we checked on November 15, 2026. It was created on August 17, 2026, so most of that attention arrived in about three months.

What OpenBot 1.0 actually is

OpenBot is a workspace for AI coworkers, which the project calls Bots. You give a Bot a task in plain words, watch it work in its own browser, review what it found, and approve the next step in the same conversation. The Bot's screen is a real browser you can take over at any point, which matters when it hits a login or two-factor screen, and you can watch its shell live while it works.

The design idea is that the workspace is modular. You bring the agents your engineers already built, or start with OpenBot's built-in Bots. You pick the models, the tools, and the infrastructure they run on. Nothing in the setup locks you to one provider's model or one company's cloud, which is the point CopilotKit keeps making against its closed rivals.

Three parts make up the system. The OpenBot backend runs the workspace API and the built-in Bots, connects external agents and tools, manages Bot computers, and enforces sign-in, tool permissions, action policies, and audit. CopilotKit Intelligence provides durable conversations, memory, Automatic Learning, and enterprise governance services. Each Bot works on a computer of its own, with its own browser, files, and shell, using Kubernetes by default or a compatible sandbox provider.

What is new in 1.0

The 1.0 release is mostly about making the thing usable without a terminal and giving teams the controls they asked for. CopilotKit's own what's new post, published November 10, 2026, lists twelve changes. The ones that matter most:

  • A desktop app for Mac and Windows. It installs the whole workspace, backend included, with no terminal. Mac builds cover Apple Silicon and Intel, and the Windows app uses the Docker Desktop you already have. Updates are signed, back up your data first, and roll back if anything fails.
  • Sign in with your AI subscription. If you have a Claude, ChatGPT, or xAI subscription, you can sign in with it and your Bots use that plan. Gemini, Unbiased AI, and any OpenAI-compatible endpoint (Azure, Bedrock, Mistral, DeepSeek, Ollama, vLLM, or your own) are also supported. You can switch the model and thinking effort for each conversation from the composer.
  • Skins for OpenDuet, OpenDots, and OpenFuse. CopilotKit previously open-sourced its own versions of Claude Duet, OpenAI dots, and Meta Muse. All three now live inside OpenBot as skins.
  • Any AG-UI agent can join as a Bot. Agents built with the Claude Agent SDK, OpenAI Agents SDK, LangChain, Google ADK, Google Antigravity, Mastra, Microsoft Agent Framework, Pydantic AI, Agno, LlamaIndex, AWS Strands, and Claude Managed Agents can plug in. Paste the agent's URL and it joins your workspace.
  • Teach a Bot by showing it once. Record a task while you do it in OpenBot's browser, talking through the steps as you go. Automatic Learning turns the recording into a proposed skill, and once you review and approve it, any Bot can perform the same task.
  • Bots that keep working in the background. Every Bot can split a big task across up to five subagents, each working on its own part at the same time. Bots can check in on things that keep changing, like an inbox or a shared document, and routines can be set up in plain words, such as a briefing on what shipped in AI agents every Monday at 8am.
  • Shared or isolated computers. All of one person's Bots share one computer by default. With OpenBot for Business, you can mark a Bot as isolated to give it a computer of its own. On the desktop app, you can also give each Bot read-only access to folders on your computer, and it asks first before editing anything.
  • Generative UI on A2UI 1.0. Bots build interfaces from a standard catalog of 65 components across layout, inputs, charts, and media. Replies stay in Markdown by default, and components appear only when they make the answer easier to read.
  • Hundreds of apps through Composio and Arcade. Connections is now one place for every app your Bots can use. Parallel Search is on from the start, Exa and Firecrawl are one click away for web search, and Notion is ready to connect. You can connect more than one account for the same app, like work and personal Gmail.
  • Clearer approvals, safe by default. Bots always ask before they delete, send, share, or post anything, even inside routines, and paying is always left to you. Approval cards say in plain words what the Bot wants to do and from which account. You can set your own approval rules for actions you trust.
  • Business controls. OpenBot for Business adds company sign-in with SSO and SCIM, boundaries that stop Bots from ever taking ruled-out actions, a full audit trail exportable as CSV, company-wide controls over approved models and MCP servers, and deployment on your own Kubernetes cluster.

How to set it up

The quickest path is the desktop app. Download the installer for Mac (Apple Silicon or Intel) or Windows from open.bot, run it, sign in, and connect your AI. CopilotKit says you are talking to your first Bot in minutes, and your first Bot comes with starter tasks, like connecting an app or searching the web, so you can see what it does right away. The app runs the backend on your laptop, and your workspace data and API keys stay on your machine.

Two things the setup needs that the headline does not mention. First, the desktop app runs its backend in Docker, so you need Docker Desktop installed, or Podman as an alternative. Second, the app depends on CopilotKit Intelligence, which needs a free local license for personal use. The desktop setup handles that for you, and the license lasts 30 days and can be renewed with a command as often as needed while your account is in good standing. Initial downloads, account sign-in, and local license renewal need internet access, even though the workspace itself runs locally.

For teams, the path is a Helm chart that deploys OpenBot and Intelligence together on AWS EKS, Azure AKS, Google GKE, or your own Kubernetes cluster. You need amd64 nodes, a default StorageClass, kubectl, Helm, a trial or production Intelligence license, and a model key. CopilotKit also offers forward-deployed engineers to help teams roll it out, and a fixed-scope proof of concept built around one workflow.

If you want to build from source instead, the repository's setup uses Docker for PostgreSQL, the API server, and the shipped Bots, plus Bun 1.4 or later for the app and setup helpers. The README notes that some tests fail on Bun 1.3, so the version floor is real. There is also a setup prompt on open.bot that you can hand to your coding agent.

What it costs

OpenBot is free for personal projects. That covers the desktop app on your laptop with Docker, your own agents, models, and tools, and a customized personal workspace. Your model provider charges separately for usage, and the pricing page is explicit that personal OpenBot is for development and experimentation. Commercial use across an organization needs OpenBot for Business.

The business plan is pay-as-you-go, at up to $29 per active user per month (about £22, about CA$41 at November 16, 2026 market rates). The per-user rate falls as the team grows: the page's own calculator prices 100 active users at $20.75 each, or $2,075 a month (about £1,569, about CA$2,967). There is no per-Bot, per-agent, or per-interaction charge, and no commitment on the monthly plan. Credit card billing is monthly.

PlanPriceWho it is forWhat you pay on top
PersonalFreeDevelopment and experimentation on your laptopYour model provider's usage, billed directly by them
Business, pay as you goUp to $29 per active user per month (about £22, about CA$41)Teams, self-hosted or hosted in early access5% of monthly model spend above $5,000 per organization; your own infrastructure
Business, AnnualCustom pricingLarger teams needing SSO, air-gapped deployment, and 24/7 supportSame 5% model-spend charge; implementation help agreed separately

Three details in the fine print decide what a team actually pays. First, an active user is a named person with at least one interaction with OpenBot for Business in the billing month. Bots, agents, API calls, and guests are never counted, and someone with no interactions in a month is not billed for that month. Second, the 5% model usage charge applies only to the part of monthly model spend above $5,000 per organization, and it applies to both pay-as-you-go and Annual. The pricing page works the example: $4,000 in monthly model spend adds no usage charge, while $10,000 in spend adds a $250 charge (about £189, about CA$357), calculated on the $5,000 above the threshold. Third, you pay your model providers directly. CopilotKit says it does not resell inference or mark up tokens.

The Annual plan is custom-priced and adds SSO and role-based access control, offline and air-gapped deployment, custom contract terms, 24/7 production support, invoicing and purchase orders, and AWS Marketplace private offers. Rates for larger teams are agreed on the first order. The pricing page lists a separate line for help from CopilotKit's team: agree on implementation assistance and ongoing support for your rollout, which means deployment help is a negotiated cost, not an included one.

Put together, the real monthly bill for a small team is the per-user charge plus model spend plus hosting. Ten active users at the top rate cost $290 a month before models and infrastructure. A hundred users at the calculator's rate cost $2,075 a month before models and infrastructure, and the 5% charge only starts once model spend passes $5,000 a month. The free tier costs nothing for one person, but it is a personal tier, not a free team plan.

The license catch

Here is the part the launch headlines mostly skip. The OpenBot application template is open source under the MIT license. You can fork it, modify it, and white-label it. But the application runs on CopilotKit Intelligence, which is licensed separately. The README says so in plain words, and the pricing page repeats it: the OpenBot application source is MIT-licensed, and the service terms for the complete offering are separate from that source license.

For personal use, that separation costs nothing in money. The desktop setup provisions a free local Intelligence license, and it can be renewed as often as needed. It does cost something in control: the local license lasts 30 days, needs internet access to renew, and depends on your CopilotKit account staying in good standing. If CopilotKit changes the terms of that free license, the free tier changes with it. That is a normal arrangement for this kind of project, but it is not the same as owning every layer of the stack.

For teams, the separation is priced in. A Kubernetes evaluation needs a free Intelligence trial license. Production needs a production Intelligence license beyond evaluation, and the OpenBot for Business license includes Intelligence for use with OpenBot. Building agents into your own product needs a separate Intelligence license on top. None of this is hidden, but it means the word "open source" describes the application template, not the whole product. Anyone comparing OpenBot with OpenClaw, which we covered in October, should know that OpenClaw's MIT license covers its own agent end to end, while OpenBot's MIT license covers the workspace around a separately licensed service layer.

OpenBot vs the other AI coworkers

OpenBot is not the first product to give an AI agent its own computer. The comparison that matters is against the other workspaces doing the same job, and CopilotKit's own comparison page lines OpenBot up against Claude, OpenAI dots, and OpenClaw. Here is that comparison with the figures we verified ourselves.

OpenBot 1.0OpenClawGrok Bot 1.0OpenAI dots
Code licenseMIT for the application template; Intelligence licensed separatelyMITClosedClosed
Where it runsYour laptop, your VPC, any Kubernetes clusterYour machine, VPS, or serverxAI's cloudOpenAI's cloud, one computer per dot
ModelsAny closed or open model, through your subscriptions or API keysAny model, from any provider or locallyxAI's Grok modelsGPT-6 Astra and GPT-6.1 Terra
Bring your own agentAny AG-UI agent, across 8+ frameworksBuilt around its own agent; coding tools plug inNo, xAI's agent onlyNo, OpenAI's agent only
Policy before each actionYes, CEL gateway, fail closedConfigurable tool permissions; its README warns to treat inbound messages as untrusted inputUser-set approval rulesUser-configured approvals
API accessYes, workspace API and AG-UI endpointsYes, runs on your own serverThrough xAI's platformThrough OpenAI's platform
PriceFree personal; up to $29 per active user per month for BusinessFree; you pay for models and hosting$30 per user per month, per our Grok Bot 1.0 reviewFirst dot included with eligible ChatGPT Pro and Business Premium plans, per CopilotKit's comparison

The table shows where OpenBot actually differs. Against OpenClaw, the difference is governance: OpenBot checks every action against a policy before it runs, with deny rules first and a fail-closed design, while OpenClaw leaves the guardrails to the user and its README. Against Grok Bot and dots, the difference is ownership: OpenBot runs on your infrastructure with any model, while the other two run on their vendor's cloud with their vendor's model. The trade is that OpenBot's free tier is personal-only, and its business tier is priced per active user, while OpenClaw is free for any use and you pay only for models and hosting.

One more comparison is worth making inside the open-source family. n8n Nod, which we covered on November 14, is also a free open-source personal agent with its own computer, aimed at one person automating their own work. OpenBot aims further: the same free desktop app, but with a business tier, SSO, audit, and Kubernetes deployment for teams. If you only need a personal agent, Nod is the simpler pick. If you need to roll agents out to a team with IT sign-off, OpenBot is built for that conversation.

What is still unproven

OpenBot 1.0 is a week old, and the honest gaps are easy to list. No independent test of the product exists yet. No benchmark, no reliability figure, and no named customer has been published. The policy gateway's behavior comes from CopilotKit's own documentation and launch coverage, not from a third-party security review. The claim that the gateway checks deny rules first, writes the audit row before acting, and refuses rather than opens on a broken rule is the vendor's description, and it should be treated as one until someone outside the company tests it.

The 20-agent story on open.bot is the same kind of evidence. CopilotKit describes a company that built its whole go-to-market team on OpenBot, with 20 AI agents in real roles and one executive setting the direction. It is a real story in the sense that CopilotKit published it with named roles and a named orchestrator. It is not independent evidence, and it is not a customer a buyer can call.

Availability is also narrower than the front page suggests. Web, iOS, Android, Slack, and Teams access are all in early access. The desktop app for Mac and Windows is the finished path today, and CopilotKit has not given dates for the other platforms. The 1.0.3 release on November 14, five days after 1.0.0, is also a reminder that this is a fast-moving codebase: three point releases in five days means bugs are being found and fixed in public, which is normal for a new open-source release and worth knowing before you put it on production work.

Common mistakes

Give each Bot only the tools and permissions it needs. OpenBot's governance helps, but a Bot with broad access and a permissive approval rule is still a Bot with broad access. Start with one narrow job, like triaging support tickets with approval before anything posts to Slack, and widen the permissions only after you have watched it work.

Do not treat the free tier as a team plan. Personal OpenBot is for development and experimentation, and the pricing page says so. A team that rolls it out commercially without a Business license is using it outside its terms. The per-active-user price also means the bill follows usage: someone with no interactions in a month is not billed, so a pilot with a small active group costs less than the headcount suggests.

Budget for the model spend, not just the seat price. The $29 per user is the ceiling for the workspace charge, but the model bill is separate and can pass it quickly. The 5% charge only starts above $5,000 a month in model spend, so a small team will not see it, but a team running heavy agent workloads should model the token cost before promising a budget.

Read the Intelligence license before you commit to self-hosting. The 30-day local license renews freely, but it needs internet access and a CopilotKit account in good standing. A fully air-gapped deployment needs the Annual plan, not the free tier.

Is OpenBot really free?

For personal use, yes. The desktop app is free, the application template is MIT-licensed, and the local Intelligence license is free and renewable. You still pay your model provider for usage, and you need Docker Desktop and internet access for setup and license renewal. Commercial team use needs the paid Business plan.

Is OpenBot fully open source?

No, not in the sense most people mean. The OpenBot application template is MIT-licensed, and you can fork and white-label it. But it runs on CopilotKit Intelligence, which is licensed separately: free locally for personal use, a free trial for Kubernetes evaluation, and a paid production license for teams. The Business license includes Intelligence for use with OpenBot.

What models can OpenBot use?

Almost any. You can sign in with a Claude, ChatGPT, or xAI subscription, or connect Gemini and other providers through API keys or OpenAI-compatible endpoints, including Azure, Bedrock, Mistral, DeepSeek, Ollama, and vLLM. Open models such as Kimi, DeepSeek, Qwen, Nemotron, and Mistral are supported. You can switch the model and thinking effort per conversation.

How much does OpenBot for Business cost?

Up to $29 per active user per month on pay-as-you-go, with the rate falling as the team grows. The pricing page's calculator shows 100 active users at $20.75 each, or $2,075 a month. A 5% charge applies to monthly model spend above $5,000 per organization. The Annual plan is custom-priced and adds SSO, air-gapped deployment, and 24/7 support.

Can I use OpenBot on my phone or in Slack?

Not yet in the finished product. iOS, Android, Slack, and Microsoft Teams access are all in early access. The desktop app for Mac and Windows is the way to use it today, and CopilotKit has not given dates for the other platforms.

How is OpenBot different from OpenClaw?

Both are MIT-licensed and self-hosted, and both give an agent its own computer. OpenBot adds built-in governance: a policy gateway that checks every action before it runs, approvals, and a full audit trail, plus a business tier with SSO and Kubernetes deployment. OpenClaw is a personal assistant you control over WhatsApp or Telegram, and its README warns you to treat inbound messages as untrusted input and set up sandboxing yourself.

The bottom line

OpenBot 1.0 is a strong open-source answer to the AI coworker question: free on your laptop, any model, any AG-UI agent, and governance that is designed in rather than added later. The license split is the thing to go in with your eyes open, and the product is too new for independent verification. For a person, the desktop app costs nothing but model spend. For a team, the $29 per active user ceiling and the $5,000 model-spend threshold make the bill easy to predict, which is more than most agent platforms offer a week after launch.

Sources

Currency conversions use November 16, 2026 market rates (1 USD = 0.756 GBP, 1 USD = 1.43 CAD) and are marked approximate.

17 min read
22 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more