OpenBot — изолированный запуск ИИ-агентов

★ 5.7K

OpenBot is an open-source platform (MIT) that gives each AI agent its own isolated computer: a dedicated Chromium browser in a container, its own file workspace, and only the tools explicitly granted by policy. Reach for it when you need to run multiple agents in parallel without them interfering with each other; when every agent action must be decided against a policy before it executes and recorded in an audit log afterward; when a security team or client needs a verifiable trail of exactly what an agent did and why. Built on TypeScript, it runs via Docker Compose with PostgreSQL as the data store; prerequisites are Bun 1.3+, Docker, a model key (OpenAI, Anthropic, or Google), and a CopilotKit Intelligence project (free plan available). It speaks the AG-UI open protocol, so you can bring agents built with LangGraph, Mastra, CrewAI, Pydantic AI, Google ADK, or written by hand — MCP servers are supported too. OpenBot is a template to clone and adapt, not a hosted product or published package. It focuses on governance — who acted and under what rule — rather than on browser automation mechanics; for pure browser scripting, a lighter tool fits better. Best suited for teams where auditability and access control over agents matter; overkill for a single script or quick prototype — currently Alpha, expect rough edges.