Skip to content

Self-Improving AI Agent that creates and refines its own tools, running them in a secure sandbox to expand its capabilities.

License

Notifications You must be signed in to change notification settings

bonk1t/agent-os

Repository files navigation

Agent OS Platform

Agent OS Platform is an open-source API and web application for managing LLM-driven multi-agent workflows. Building on OpenAI's Assistants API, it offers a collaborative environment for developing, testing, and deploying AI teams.

Architecture

Backend

  • FastAPI application for API and WebSocket endpoints
  • Firebase Authentication and Firestore for data persistence
  • E2B for secure sandbox execution
  • Redis for message bus and state management

Frontend

  • Gatsby-based web application
  • TailwindCSS for styling
  • Ant Design for UI components
  • Real-time updates via WebSocket

Key Features

  • Configuration Management: Centrally manage configurations for agencies, agents, and skills
  • Custom Skills: Extend AI agents with specialized skills
  • Secure Execution: Isolated sandbox environments for running agent code
  • Real-time Communication: WebSocket support for live updates
  • Modern UI: Beautiful and responsive interface with best UX practices

Getting Started

Quick Start with Docker

  1. Create .env.docker from .env.docker.testing template
  2. Run:
    source .env.docker
    FIREBASE_CONFIG=$FIREBASE_CONFIG docker-compose up --build

Local Development

  1. Backend (FastAPI):

    cd backend
    pip install -r requirements.txt
    uvicorn main:app --reload
  2. Frontend (Gatsby):

    cd frontend
    npm install
    yarn start

For detailed setup instructions, refer to:

Status

Note

Agent OS Platform is a research project exploring multi-agent workflows.

About

Self-Improving AI Agent that creates and refines its own tools, running them in a secure sandbox to expand its capabilities.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published