OpenManus is an open-source framework for building general AI agents, designed to empower users to create their own agents without invite codes. It targets developers and researchers interested in agent development, offering a flexible platform for rapid prototyping and experimentation.

How It Works

OpenManus provides a foundational structure for AI agents, allowing users to define their ideas and have the framework orchestrate the necessary steps. It supports various operational modes, including a core agent, a Multi-Agent Collaboration Protocol (MCP) tool version, and an unstable multi-agent version. The project also introduces OpenManus-RL for reinforcement learning-based tuning of LLM agents.

Quick Start & Requirements

• Installation: Recommended via uv (install with curl -LsSf https://astral.sh/uv/install.sh | sh), then uv venv --python 3.12, source .venv/bin/activate, and uv pip install -r requirements.txt. Conda installation is also supported.
• Prerequisites: Python 3.12, playwright for browser automation (optional).
• Configuration: Requires a config/config.toml file with LLM API keys (e.g., OpenAI).
• Running: python main.py for the core agent, python run_mcp.py for MCP, python run_flow.py for multi-agent.
• Links: Project Demo, Zenodo Citation

Highlighted Details

• Prototype launched within 3 hours.
• Includes OpenManus-RL for reinforcement learning-based LLM agent tuning.
• Supports browser automation via Playwright.
• Developed by contributors from the MetaGPT community.

Maintenance & Community

The project is actively developed by core authors from MetaGPT and welcomes contributions. A Feishu group is available for networking.

Licensing & Compatibility

The project is available under a permissive license, facilitating commercial use and integration into closed-source projects.

Limitations & Caveats

The multi-agent version is noted as unstable. Configuration requires manual setup of LLM API keys.