OpenManus is an open-source Python framework for building general-purpose AI agents, created by contributors from the MetaGPT team as a freely usable alternative to the invite-gated Manus agent. You run it from the terminal, type an idea, and the agent executes it; three entry points are provided — the standard single-agent runner, an MCP tool version, and an experimental (described as unstable) multi-agent "flow" mode that can coordinate multiple agents, including an optional DataAnalysis agent for data analysis and visualization tasks.
Configuration is a TOML file pointing at any OpenAI-compatible LLM endpoint, with a separate optional vision-model entry. Browser automation is available as an optional capability via Playwright, and the project builds on foundations from anthropic-computer-use, browser-use, and crawl4ai. A companion project, OpenManus-RL, explores reinforcement-learning-based tuning methods (such as GRPO) for LLM agents in collaboration with UIUC researchers.
