OpenManus is an open-source project developed by a team from MetaGPT, designed to enable users to build and customize AI agents without the need for an invite code. It offers a simple yet powerful implementation, making it accessible for developers to contribute and extend its capabilities. The project includes OpenManus-RL, a specialized module focusing on reinforcement learning (RL)-based tuning methods for LLM agents, developed in collaboration with researchers from UIUC.
Key Features:
- Easy Installation: Supports two installation methods, with UV recommended for faster setup and better dependency management.
- Configurable LLM APIs: Users can easily configure the framework to work with various LLM models, including GPT-4o, by editing a simple
config.tomlfile. - Multiple Execution Modes: Offers different scripts for running the agent, including a basic version (
main.py), an MCP tool version (run_mcp.py), and an experimental multi-agent version (run_flow.py). - Community-Driven: Encourages contributions and feedback, with pre-commit checks to ensure code quality.
Installation:
- Using Conda:
bash conda create -n open_manus python=3.12 conda activate open_manus git clone https://github.com/mannaandpoem/OpenManus.git cd OpenManus pip install -r requirements.txt - Using UV:
bash curl -LsSf https://astral.sh/uv/install.sh | sh git clone https://github.com/mannaandpoem/OpenManus.git cd OpenManus uv venv --python 3.12 source .venv/bin/activate # On Unix/macOS uv pip install -r requirements.txt playwright install
Configuration:
After installation, users need to set up their LLM API keys in the config.toml file:
toml
[llm]
model = "gpt-4o"
base_url = "https://api.openai.com/v1"
api_key = "sk-..." # Replace with your actual API key
max_tokens = 4096
temperature = 0.0
Usage:
Run the agent with:
bash
python main.py
Then input your idea via the terminal!
Contributions:
OpenManus welcomes contributions from the community. Users can submit issues or pull requests, or contact the team via email at mannaandpoem@gmail.com.
Acknowledgements:
The project is supported by PPIO for computing resources and acknowledges contributions from anthropic-computer-use, browser-use, AAAJ, MetaGPT, OpenHands, SWE-agent, and stepfun (阶跃星辰) for Hugging Face demo space support.
For more details, visit the GitHub repository.