CopilotKit Releases OpenBot 1.0: An AI Coworker That Audits Every Action Before It Runs
CopilotKit has released OpenBot 1.0, an open-source AI coworker that companies run on their own infrastructure. Before each click, file write, or tool call, a governance gateway checks the action against company policy, and the audit log entry is written before the action executes. If a rule is violated or broken, the action is blocked rather than allowed through. For businesses, this is a way to give an agent real permissions without giving up control of the stack.
AI-processed from X: @Marktechpost; edited by Hamidun News
The main reason companies are cautious about AI agents is simple: an agent that can click, write files, and call external services can just as easily do something it shouldn't. Usually it either gets no real permissions, or it gets them and everyone hopes for the best. CopilotKit offers a third path. It has released OpenBot 1.0, an open AI coworker that an organization deploys on its own infrastructure and constrains with its own rules.
A Gateway That Says "No" by Default
The central idea of the release is the governance gateway. It sits between the agent (the project calls it Bot) and everything the agent does. Every click, every file write, and every tool call is first checked against company policy. The action runs only after the check passes.
The gateway has two properties that matter to both security teams and managers.
First, the audit line is written before the action is executed. The log doesn't depend on how things turned out. If the agent hangs or crashes halfway through, the log still contains a record of what it was about to do. This matters enormously during incident review: you see the intent, not just the result.
Second, a broken rule leads to a refusal, not a pass-through. In security this is called fail-closed. Think of an office turnstile: if the badge system fails, it stays shut rather than letting everyone through. The same applies here. If the policy hits an error, the agent gets no access.
Your Own Stack and Nine Model Families
OpenBot's second argument is the absence of vendor lock-in. You can connect any agent that speaks the AG-UI protocol. The list of compatible frameworks includes:
- Claude Agent SDK and OpenAI Agents SDK;
- LangChain and CrewAI;
- Google ADK and Mastra;
- AWS Strands and Microsoft Agent Framework.
Models aren't limited to a single vendor either: nine families are supported. These are OpenAI, Claude, Gemini, and Grok, plus Kimi, DeepSeek, Qwen, Nemotron, and Mistral. For a team, this means an agent that's already written doesn't have to be rebuilt for a new platform, and the model can be swapped without touching the safety rules. Policies and auditing live in the gateway, not in the agent itself.
Deployment is described as a single Helm chart. It can be rolled out to AWS, Azure, GCP, or any Kubernetes cluster. Data and agent actions stay inside your perimeter, which for banks, healthcare providers, and companies with strict data-residency requirements is often the deciding argument.
License, Price, and What Isn't Available Yet
The code is released under the MIT license. OpenBot is free for personal use, and for business the price goes up to $29 per active user per month. There is a trade-off to weigh in advance. SSO, SCIM, and token limits are placed in the paid business tier. That means single sign-on, automated account management, and control over model spending are not available in the free version, and for an enterprise rollout these are exactly the baseline features. In addition, the iOS and Android mobile apps and the Slack and Teams integrations are still in early access. If your employees work in messengers, this is not a ready-made solution on launch day.
What to Do in Practice
If you're building an internal agent or planning a pilot, OpenBot is worth a look at minimum as an architectural reference: policy checks before action, audit before execution, and refusal on failure. You can adopt these three principles even if you never use the product itself.
For product managers and CTOs, a useful question is: what exactly are we prepared to allow the agent to do, and who will record that in writing? The gateway turns the answer into rules that can be verified and revised. But how well those rules hold up in real scenarios can't be judged from a single announcement; you need your own tests. Before scaling a pilot, check whether the free features are enough for you, or whether you'll be paying for SSO and token limits from day one.
- Nov 21, 2026Plural Raises €1 Billion, Double Its Last Fund, to Back European Startups
- Nov 21, 2026Nvidia Board Ousts Jensen Huang, Putting $12.9B Hugging Face Deal at Risk
- Nov 21, 2026OpenAI Lets Four External Auditors Inside Its Models: Apollo, METR, Transluce, and Goodfire
- Nov 21, 2026California Sues OpenAI Over AI Agents That Hacked Hugging Face: $2,500 Fine per Violation
Key Questions Answered
What are the key details on A Gateway That Says "No" by Default?
The central idea of the release is the governance gateway. It sits between the agent (the project calls it Bot) and everything the agent does. Every click, every file write, and every tool call is first checked against company policy. The action runs only after the check passes. The gateway has two properties that matter to both security teams and managers. First, the audit line is written before the action is executed. The log doesn't depend on how things turned out.
What are the key details on Your Own Stack and Nine Model Families?
OpenBot's second argument is the absence of vendor lock-in. You can connect any agent that speaks the AG-UI protocol. The list of compatible frameworks includes: Claude Agent SDK and OpenAI Agents SDK; LangChain and CrewAI; Google ADK and Mastra; AWS Strands and Microsoft Agent Framework. Models aren't limited to a single vendor either: nine families are supported. These are OpenAI, Claude, Gemini, and Grok, plus Kimi, DeepSeek, Qwen, Nemotron, and Mistral.
What are the key details on License, Price, and What Isn't Available Yet?
The code is released under the MIT license. OpenBot is free for personal use, and for business the price goes up to $29 per active user per month. There is a trade-off to weigh in advance. SSO, SCIM, and token limits are placed in the paid business tier. That means single sign-on, automated account management, and control over model spending are not available in the free version, and for an enterprise rollout these are exactly the baseline features.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.