Verified project record

CopilotKit/OpenBot

OpenBot is a self-hosted AI coworker platform where each agent (Bot) runs in its own container with its own browser profile and workspace, and every action is checked against policy and audited before it happens. It is an alpha-stage template to clone and run yourself, not a hosted sign-up product.

Project overview

It combines per-Bot sandboxed containers with policy-checked, audited actions delivered over the framework-agnostic AG-UI protocol, so governance is not tied to a specific agent framework.

Project type: MCP · AI Agent
Use cases: Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers and AI engineers who want a self-hosted template to clone and run, with per-agent sandboxing and pre-action policy enforcement over the AG-UI protocol.
Teams that need to own and customize AI assistants with tools inside their own infrastructure and require every agent action to be policy-checked and audited.

Key capabilities

Any AG-UI endpoint can be a Bot, so OpenBot is not tied to an agent framework.
The supervisor gives each Bot its own container, its own /workspace volume, and its own browser profile.

Limitations and risks

Expect rough edges, bugs, and moving parts.
The agent-computer drives a browser holding real logins; it must stay bound to loopback.
A fresh clone admits every request as one administrator until sign-in is configured.
A CopilotKit Intelligence project and license are required (a free plan is available; it can be self-hosted), and a model key is required — OpenAI for the POC Bot, or OpenAI, Anthropic, or Google for the LangGraph Bot. No model ships in the box. The overall cost dependency is unclear.
GPU requirements, telemetry behavior, and the end-to-end data boundary are not documented. While local orchestration and local deployment are documented, inference depends on an externally supplied model API key and on a CopilotKit Intelligence project unless that project is self-hosted, so a local-only data path is not established by the documented facts. The coding requirement is unclear.

Getting started

Copy .env.example to .env and run bun install; connect an Intelligence project (npx copilotkit login and project select, then bun scripts/setup-learning.ts); set OPENAI_API_KEY; run bash scripts/start.sh; open http://localhost:3010. Setup difficulty is medium, requiring filled .env keys, an Intelligence project, and a model API key.

Alternatives and comparisons

Listed as a candidate alternative for local, privacy-oriented agent browser use; compare directly against OpenBot's per-Bot sandboxed, policy-governed platform before choosing.
Listed as a candidate alternative single self-hostable agent runtime; compare against OpenBot's per-Bot sandboxed containers and pre-action policy enforcement before choosing.
Listed as a candidate alternative for agent-driven computer and browser control; compare against OpenBot's governance layer before choosing.

Project comparisons

Evidence and sources

README: A template, not a product. OpenBot is meant to be cloned and made your own. There is no hosted version to sign up for, and nothing here is published as a package to depend on
README: it runs on your infrastructure and you can change anything about it. Any agent stack, through AG-UI.
README: An agent platform that runs inside your own infrastructure. Docker Compose brings up every part of it, the data sits in your PostgreSQL, and the model is yours to choose
README: Alpha, and under active development. OpenBot is early. Expect rough edges and bugs
README: A computer per Bot: the supervisor gives each Bot its own container, its own /workspace volume and its own browser profile.

Source updated at: 2026-09-29T18:31:50Z

Open original GitHub project