OpenManus just killed the invite wall for AI agents. Here's how to run it in ten minutes.

If you have been watching the AI agent space with a mix of fascination and frustration, fascinated by what tools like Manus can do, frustrated that you cannot get an invite code, the wait is over. OpenManus, released by the team behind MetaGPT, is here, and it is designed to be as open and accessible as possible.

The prototype was built in three hours by a team including Xinbin Liang, Jinyu Xiang, Zhaoyang Yu, Jiayi Zhang, and Sirui Hong. That speed is not sloppiness; it reflects the maturity of the underlying components. OpenManus borrows from proven projects like Anthropic's computer-use, browser-use, and crawl4ai, and wraps them into a clean, extensible framework.

Installation is trivially simple

Getting started takes minutes. The recommended method uses uv, a fast Python package installer:

curl -LsSf https://astral.sh/uv/install.sh | sh
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

Then copy the example config file:

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

Edit config/config.toml to add your preferred LLM's API key, OpenAI's GPT-4o is used in the example, but any compatible model works, and you are ready.

A single command launches the agent:

python main.py

Then type your idea directly into the terminal. No dashboard, no cloud subscription, no approval queue.

Not just a single agent

OpenManus ships with multiple execution modes. The basic main.py runs a general-purpose agent. For multi-agent orchestration, run_flow.py is available, though the team labels it as unstable. A dedicated data analysis agent can be activated in config.toml by setting use_data_analysis_agent = true in the [runflow] section.

A browser automation tool is optional but recommended for agents that need to navigate the web. Install it with:

playwright install

The RL frontier: OpenManus-RL

Beyond the immediate agent framework, the project introduces OpenManus-RL, a companion effort focused on reinforcement learning-based tuning methods, specifically Group Relative Policy Optimization (GRPO), for LLM agents. Developed in collaboration with researchers from UIUC, this is aimed at those who want to push agent performance beyond prompt engineering.

It is a simple implementation, so we welcome any suggestions, contributions, and feedback.

The team is candid about the project's early stage, but that is exactly the point: the community can shape it from day one.

Why this matters right now

The AI agent landscape has been dominated by proprietary, invite-only systems. Manus demonstrated what was possible, but kept the keys locked. OpenManus breaks that model. Any developer with a Python environment and an API key can now run autonomous agents, integrate them into workflows, and experiment with multi-agent architectures, all without asking for permission.

For startups, researchers, and hobbyists, this lowers the barrier to entry to virtually zero. The project also accepts pull requests and issues, meaning the next innovation could come from anyone.

Sponsorship and backing

Computing support for the project comes from PPIO, a GPU cloud provider positioned as an affordable and easily integrated MaaS solution. The Hugging Face demo space is supported by stepfun (阶跃星辰).

The code is available under a Zenodo citation (doi: 10.5281/zenodo.15186407) and can be cited in academic work.

What to do next

OpenManus is ready. The installation steps above will get you a running agent in under ten minutes. The repository is at github.com/FoundationAgents/OpenManus. There is no reason to wait, the only thing missing is your idea.