OpenBot is an open-source AI agent platform that runs on your own infrastructure, giving each AI coworker its own isolated browser, file system, and access to only the tools you explicitly grant.
The platform solves the problem of deploying AI agents safely in production by implementing a governance layer that sits between agents and their actions. Every action an agent takes—whether on a computer, file system, MCP server, or UI component—passes through a central gateway that decides whether to allow it before execution and records it afterward. This approach lets you hand real work to AI agents while maintaining visibility and control over what they do. The system runs entirely on your machine using Docker Compose, stores data in your own PostgreSQL database, and requires you to supply your own model credentials, which are encrypted at rest and never logged.
OpenBot is designed as a template to clone and customize rather than a hosted service. It ships with three example coworkers—a General Assistant, Knowledge agent, and Risk Analyst—configured through YAML rather than code, making it straightforward to add your own agents. The platform is built on AG-UI, an open protocol for agent-to-user interaction, which means you can bring agents built with LangGraph, Mastra, CrewAI, Pydantic AI, Google ADK, or custom implementations. This framework-agnostic approach ensures governance applies consistently regardless of how the underlying agent was built. The tool suits teams that need to deploy AI agents with strong access controls and audit trails, particularly in regulated environments or where agents interact with sensitive systems.
Development is active but early-stage, with the project explicitly marked as alpha and under active development. The codebase is written in TypeScript and all workspaces are private, requiring users to fork and customize the repository for their own use. The maintainers welcome issues and pull requests as the project evolves.
