OpenManus

OpenManus is an open-source framework for building general AI agents, developed by the MetaGPT team at FoundationAgents. The project was created to provide an open alternative to proprietary agent systems, demonstrating that sophisticated AI agent capabilities can be achieved without requiring invite codes or closed platforms.

OpenManus supports tool use, web browsing, file operations, code execution, multi-agent collaboration, and browser automation. It is designed to be extensible and modular, allowing developers to add custom tools and agent behaviors. The project is MIT-licensed and hosted at github.com/FoundationAgents/OpenManus.

OpenManus requires Python 3.12 and can be installed via conda or uv.

Using uv (Recommended)

git clone https://github.com/FoundationAgents/OpenManus.git
cd OpenManus
uv venv --python 3.12
source .venv/bin/activate
uv pip install -r requirements.txt

Using conda

conda create -n open_manus python=3.12
conda activate open_manus
git clone https://github.com/FoundationAgents/OpenManus.git
cd OpenManus
pip install -r requirements.txt

Browser Automation (Optional)

playwright install

OpenManus requires configuration for the LLM APIs it uses. Create a config.toml file in the config directory with your API settings:

cp config/config.example.toml config/config.toml

Edit config/config.toml to add your API keys and customize settings for the LLM model, base URL, max tokens, temperature, and optional vision model configuration.

Run OpenManus with a single command from the terminal:

python main.py

Then input your idea via terminal. The agent will process your request using available tools. For MCP tool support, run python run_mcp.py. For the multi-agent version with multiple collaborating agents, run python run_flow.py.

OpenManus also includes a specialized DataAnalysis Agent for data analysis and visualization tasks. Enable it in config.toml with runflow.use_data_analysis_agent = true.

• Tool Use — Agents can use tools including web search, file operations, code execution, and browser automation
• Browser Automation — Playwright-based web browsing for data extraction and web interaction
• Multi-Agent Collaboration — Run multiple agents that work together on complex tasks via run_flow.py
• MCP Support — Connect to MCP servers for extended tool capabilities
• Data Analysis — Built-in DataAnalysis agent for charts, statistics, and data processing
• Extensible Tool System — Add custom tools and agent behaviors through a modular architecture
• Vision Support — Optional vision model configuration for image-based tasks

OpenManus supports running multiple agents in a collaborative flow architecture. The multi-agent system allows different specialized agents to work together on complex tasks, passing results between each other. The project also includes OpenManus-RL, a reinforcement learning extension for tuning LLM agents using GRPO-based methods, developed in collaboration with UIUC researchers.

The agent architecture is inspired by Anthropic's computer-use demo, browser-use, and crawl4ai. It is designed to be modular and extensible, with a clear separation between the agent logic, tool implementations, and LLM backend. The flow system allows you to define custom agent pipelines by composing different agent types in a directed graph.