Skip to content
Get packs

Guide · Open source

How to use OpenManus: install, configure and run it

Updated · written with AI from the project's own documentation and checked against our catalogue data

OpenManus is an open-source Python project for building AI agents, programs that take a task in plain language and use a language model to work toward it. It is aimed at developers, students and teams who want to run and customize a general agent themselves. This guide walks through the prerequisites, installation with uv or conda, configuration with your own API key, and the main ways to run it.

OpenManusFoundationAgents/OpenManus · ★ 58.9k · MIT

Quick start

  1. 1Check prerequisites. Confirm you have Python 3.12, Git, and uv or conda, plus an LLM API key.
  2. 2Clone the repository. Run git clone on the FoundationAgents/OpenManus repository and enter the folder.
  3. 3Create the environment. Create a virtual environment with uv venv --python 3.12 and activate it.
  4. 4Install dependencies. Run uv pip install -r requirements.txt inside the activated environment.
  5. 5Add your API key. Copy config/config.example.toml to config/config.toml and set your api_key and model.
  6. 6Run the agent. Start python main.py and type your task into the terminal.

What you need before you install OpenManus

You need Python 3.12, Git, and either uv or conda. Either package manager works according to the README. You also need an LLM API key; the README example uses OpenAI, and you supply your own key.

  • Python 3.12
  • Git
  • uv or conda
  • An LLM API key (the README example uses OpenAI)

Install OpenManus with uv (recommended)

The README recommends the uv method for faster installation and better dependency management. Clone the repository, create a virtual environment, then install the dependencies from requirements.txt.

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

Install OpenManus with conda instead

If you prefer conda, create an environment with Python 3.12, clone the repository, and install the same requirements file with pip.

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

Add your LLM API key in config.toml

OpenManus needs configuration for the LLM APIs it uses. Copy the example file to config/config.toml, then edit it to add your API key and model settings. The README example uses gpt-4o with the OpenAI base URL and a max_tokens of 4096.

Keep your key private: the config file holds it in plain text, and the example contains a placeholder that you must replace.

  • Copy config/config.example.toml to config/config.toml
  • Set model, base_url and api_key under [llm]
  • Optionally set a separate [llm.vision] block
cp config/config.example.toml config/config.toml # config/config.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

Run OpenManus and give it a task

Run the main script, then type your idea into the terminal. The README also lists two other entry points: run_mcp.py for the MCP tool version and run_flow.py for a multi-agent version that the README describes as unstable.

To enable the DataAnalysis agent, set use_data_analysis_agent to true under [runflow] in config.toml. It is disabled by default, and the README says it needs extra dependencies from its own installation guide.

  • python main.py: the standard single-agent run
  • python run_mcp.py: MCP tool version
  • python run_flow.py: multi-agent version (marked unstable in the README)
python main.py python run_mcp.py python run_flow.py

Enable browser tasks with Browser Use

OpenManus starts Browser Use CLI 3.0 as a default MCP server, so the agent can use a browser. Local mode attaches to Chrome or Chromium automatically and needs no API key. Run the doctor command to check your setup or install Chromium.

To turn the default server off, set OPENMANUS_DISABLE_BROWSER_USE=1. BrowserGym is a separate option and still needs its Playwright browser.

  • Local mode: attaches to Chrome or Chromium, no API key needed
  • Cloud mode: set BROWSER_USE_API_KEY, or select a browser with BU_CDP_URL, BU_CDP_WS or BU_NAME
  • Disable the default server with OPENMANUS_DISABLE_BROWSER_USE=1
uvx browser-use --doctor uvx browser-use install playwright install

What people use it for

  • Experimenting with how an LLM agent calls tools and browses the web, as a learning project.
  • Running a general agent that takes a plain-language task and works toward it from the terminal.
  • Customizing agent behavior in Python instead of relying on a closed product.
  • Running a data analysis agent for analysis and visualization tasks by enabling it in config.toml.

Frequently asked questions

Is OpenManus free to use?

OpenManus is released under the MIT licence. The README, however, requires an API key for the LLM models it calls, so you pay for that model usage with your provider.

What do I need to install OpenManus?

The README lists Python 3.12, Git, uv or conda, and an LLM API key. The example uses OpenAI, but you supply your own key.

Can I use OpenManus without an OpenAI key?

The README's example configuration uses OpenAI, and the config file takes a base_url and api_key for an LLM. Compatible providers are mentioned in the project description, but check the config example for the exact fields.

How do I fix a configuration error in OpenManus?

Make sure you copied config/config.example.toml to config/config.toml and replaced the placeholder api_key with your real key. The README does not list other troubleshooting steps.

How does OpenManus compare to AutoGPT or MetaGPT?

AutoGPT, OpenHands, MetaGPT and SWE-agent are listed as comparable projects. The README credits MetaGPT and OpenHands among its acknowledgements, but it does not provide a benchmark comparison, so test them on your own tasks.

Is OpenManus ready for production use?

The README calls its implementation a simple one that is still evolving and welcomes feedback. It is better suited to experimentation and learning than to a supported production deployment.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

OpenManus suits developers, students and teams who want an open, Python-based general agent they can read and change, and who already have an LLM API key. It is a poor fit for non-technical users or anyone who needs a polished, supported production product, since the README describes it as simple and still evolving.

Install options & downloads