CopilotKit/OpenBot

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

What it solves

OpenBot provides a secure, governed environment for deploying AI agents as "coworkers" that can perform real-world tasks on a computer. It solves the trust and security gap when giving agents access to browsers, files, and system tools by routing every action through a strict governance gateway that records and validates every move before it happens.

How it works

OpenBot acts as an agent platform that runs on your own infrastructure. It uses the AG-UI open protocol, allowing it to integrate with agents built from any framework (like LangGraph, CrewAI, or Pydantic AI).

Key technical components include:

  • Isolated Environments: Each bot is assigned its own Docker container with a dedicated browser profile, workspace volume, and shell.
  • Governance Gateway: All tool calls (browser, file, or MCP) must pass through a central gateway that evaluates a CEL-based policy (fail-closed) and writes an audit log before execution.
  • Human-in-the-Loop: Users can "take the wheel" to handle 2FA or login walls, pausing the bot's actions until control is released.
  • Component-Based UI: Bots communicate using React components rather than just text, allowing for richer interactions.

Who it’s for

It is designed for developers and organizations that need AI agents to perform autonomous work with sensitive access (logins, files) but require strict audit trails, security boundaries, and the ability to intervene manually.

Highlights

  • Isolated Bot Computers: Every bot gets its own containerized browser and filesystem.
  • Strict Audit Trail: A readable log of every permitted, refused, or failed action.
  • Framework Agnostic: Supports any agent via the AG-UI protocol.
  • Secure Secret Management: Credentials are encrypted at rest and redacted from transcripts.
  • Governed MCP: Integrated support for a curated catalog of MCP servers (e.g., Slack, Salesforce) with read/write classification.

Related

  • Project
    obot-platform/obot

    An open-source platform for organizations to manage, secure, and govern their AI ecosystems, providing centralized control over LLM access, MCP servers, and AI activity auditing.

  • Project
    Kritt-ai/open-kritt

    An open-source security research platform that orchestrates AI agents to find, validate, and rank vulnerabilities in codebases using focused, parallelized workflows.

  • Project
    AIDotNet/OpenCowork

    OpenCowork is an open‑source desktop (Electron + .NET) platform that lets LLM‑based agents work directly on your local files, shell, and repositories. It supports multi‑agent teams, integrates with many chat services, offers five interaction modes (chat, clarify, cowork, code, acp), and provides a rich toolbox (file/shell access, browser automation, task/cron management, CodeGraph indexing, skill extensions). The same native .NET worker powers both the GUI and a terminal‑based CLI, keeping a single authoritative agent loop. Quick start requires Node ≥ 18 and .NET 11 SDK; builds are available for Windows, macOS, and Linux. Use cases include autonomous coding, scheduled ops, research/report generation, remote SSH tasks, and media‑pipeline design. Licensed under Apache 2.0.

  • Project
    opensandbox-group/OpenSandbox

    A general-purpose sandbox platform for AI applications that provides secure, isolated environments for code execution, agent evaluation, and RL training.

  • Project
    microsoft/agent-governance-toolkit

    A governance framework for autonomous AI agents that provides deterministic policy enforcement, identity management, and audit logging to prevent unauthorized actions.