OpenManus
Python general-purpose AI agent you run from a terminal with your own LLM API key, with MCP and browser tools.
Give it an idea in the terminal; it works through your own LLM API key and browser tools.
OpenManus is an open-source general-purpose AI agent written in Python, built as an invite-free alternative to Manus. You run it from a terminal against an LLM API you configure. It suits developers who want to experiment with agents rather than buy a finished product.
What it is
You install the Python dependencies, add your LLM endpoint and API key to config.toml, and run python main.py to type in a task. There is an MCP tool version (run_mcp.py), an unstable multi-agent flow (run_flow.py), an optional data-analysis agent, and a Browser Use CLI wired in as the default MCP server for browser automation.
Who it's for
It suits developers and researchers who are comfortable with Python, conda or uv, and API keys, and want to read and modify an agent. It does not suit non-technical users wanting a polished app: the README calls the multi-agent version unstable and the project describes itself as a simple implementation.
Key features
Terminal-driven agent
Run python main.py and enter your idea at the prompt.
MCP tool version
run_mcp.py runs the agent with Model Context Protocol tools.
Browser Use as default MCP
Browser automation runs through uvx browser-use, attaching to local Chrome or a cloud browser.
Optional data analysis agent
Enable a DataAnalysis agent in run_flow for charts and data analysis.
Configurable LLM endpoint
Set model, base URL and API key in config.toml, with a separate vision model block.
Multi-agent flow
run_flow.py offers a multi-agent mode the README labels unstable.
Up and running in 3 steps
Our honest verdict
What's great
- You choose the LLM and endpoint in config.toml, including a separate vision model.
- Browser automation is built in through Browser Use, with a local mode that needs no API key.
- An MCP version lets you plug the agent into the tool ecosystem.
- The code is small enough, by its own description a simple implementation, to read and modify.
Where it falls short
- You pay your own LLM API costs; the example config uses a gpt-4o key and the agent has no model of its own.
- Setup means conda or uv, Python 3.12, Playwright browsers and a hand-edited config.toml, which takes technical skill.
- The README labels the multi-agent flow unstable, so expect rough edges on complex tasks.
- Results depend on the model you connect; the README gives no accuracy figures or task success rates.
Pricing & license
OpenManus is MIT-licensed and free to run. The README shows no paid plan, so the real cost is the LLM API usage you pay to your model provider, plus an optional Browser Use Cloud key if you choose a remote browser.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Does OpenManus need an invite code?
No. The README says it can achieve ideas without an invite code, unlike the original Manus.
Which models can it use?
You set the model, base URL and API key in config.toml; the example uses gpt-4o with the OpenAI endpoint.
Can it control a browser?
Yes. It starts Browser Use CLI as a default MCP server, attaching to Chrome or Chromium locally or to a Browser Use Cloud browser.
