OpenManus

Built in roughly three hours as an open replication of the invite-only Manus agent, OpenManus grew into a general-purpose autonomous agent framework maintained by FoundationAgents, a team drawn from the MetaGPT project. An agent run starts with python main.py: the agent plans a task, then executes it using browser automation through Playwright, code execution, file operations, and any tools exposed through Model Context Protocol servers, with a run_flow mode for multi-agent workflows and a dedicated DataAnalysis agent for data visualization tasks. It requires Python 3.12 and API credentials for a configured LLM in config.toml, and any OpenAI-compatible endpoint works, with GPT-4o used in the documented examples; conda or the faster uv package manager handles setup. A companion project, OpenManus-RL, explores reinforcement learning tuning for agents. Everything is MIT licensed, and the repository passed 57,000 GitHub stars within months of its March 2025 launch, making it one of the most visible open agent projects.