Skip to content
Get packs

OpenManus

Open-source Python framework for building general-purpose AI agents that run tasks from a prompt.

GitHub Download v0.3.0
Stars
58.9k
Forks
10.2k
Licence
MIT
Language
Python
Last update
4 days ago
Type
AI model / agent

What it is

OpenManus is a Python project for building AI agents, programs that take a task described in plain language and use a language model to work toward it. The README describes it as an open alternative to the Manus agent that does not require an invite code. You supply your own LLM API key and run the agent from a terminal, and it can also use a browser through Browser Use.

  • Give the agent a task in the terminal and let it work through it step by step
  • Run the MCP version (run_mcp.py) to use tools through the Model Context Protocol
  • Use the optional DataAnalysis agent for data analysis and visualization tasks after enabling it in config.toml
  • Let the agent control a local Chrome or Chromium browser via Browser Use

Great for

  • Developers who want to experiment with an open-source general AI agent
  • Students learning how LLM agents call tools and browse the web
  • Teams wanting to customize agent behavior in Python rather than use a closed product
  • People who already have an LLM API key from OpenAI or a compatible provider

Not ideal for

  • Non-technical users, since setup requires Python, a config file and API keys
  • Anyone needing a polished, supported production product, because the README calls its implementation simple and is still evolving
  • Users unwilling to pay for LLM API usage, since the README requires an API key for the models it calls
Step-by-step guide: how to use OpenManusInstall, first run, examples, tips and FAQs
OpenManus alternativesOpenHands vs OpenManusMetaGPT vs OpenManusManus vs OpenManusOpenManus vs SWE-agent

How to get it

  1. 1

    Clone and build from source

    You'll need: Python 3.12 · Git · uv or conda (either method works) · An LLM API key (the README example uses OpenAI)

    Clone the repository

    git clone https://github.com/FoundationAgents/OpenManus.git

    Create and activate a virtual environment with uv

    uv venv --python 3.12

    Install dependencies with uv

    uv pip install -r requirements.txt

    Copy the example configuration

    cp config/config.example.toml config/config.toml
  2. 2

    Run it

    • Edit config/config.toml and add your LLM API key and model settings
    • Run python main.py from the OpenManus folder
    • Type your task into the terminal and wait for the agent to respond
    • Optionally try python run_mcp.py for the MCP tool version
  3. 3

    Get the most out of it

    • Enable the DataAnalysis agent by setting use_data_analysis_agent = true under [runflow] in config.toml, then install its dependencies per the linked guide
    • Configure a separate vision model under [llm.vision] in config.toml if you need image understanding
    • Connect the default Browser Use MCP server to an existing browser using BU_CDP_URL, BU_CDP_WS or BU_NAME
    • Set OPENMANUS_DISABLE_BROWSER_USE=1 if you do not want the default browser tooling

Downloads

Source code (.zip)main branch · straight from GitHubRelease v0.3.0Release notes and sourceOfficial website & docsopenmanus.github.io/

Strengths

  • MIT licensed, so the code can be used and modified freely
  • Supports multiple entry points: a general agent, an MCP version and a multi-agent flow
  • Clear configuration through a single config.toml file for model and API settings

Trade-offs

  • The README calls its multi-agent flow unstable and describes the project as a simple implementation
  • Requires paid or self-hosted LLM API access and careful key handling
  • Several optional features need extra setup, such as Playwright browsers and chart visualization dependencies

How it compares

ProjectStarsTypeLevelUpdatedBest for
OpenManus (this)58.9kAI model / agentintermediate4 days agoDevelopers who want to experiment with an open-source general AI agent
OpenHands92.5kAppintermediate4 days agoDevelopers who want coding agents running on a server while their laptop is off
MetaGPT71.1kFrameworkintermediate10 months agoDevelopers experimenting with multi-agent LLM workflows
SWE-agent20.7kAI model / agentintermediate10 days agoResearchers studying autonomous coding agents or running SWE-bench experiments

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.