foundationagents/openmanus

OpenManus is an open-source Python framework for building and running AI agents, software programs that can autonomously plan and execute multi-step tasks by calling tools, browsing the web, writing code, and taking other actions. It was created as an open alternative to Manus, a closed commercial AI agent that required an invite code, built by a team originally from the MetaGPT project. The core problem it addresses is giving developers a free, accessible way to run capable AI agents without waiting lists or paywalls. An "agent" in this context means a system that receives a high-level goal in plain text ("research this topic and write a report," "analyze this dataset," "build a small web app") and then breaks that goal down into steps, selects appropriate tools, executes those tools, observes the results, and continues until the task is done. OpenManus works by connecting a large language model (you configure which one, GPT-4o, Claude, or any OpenAI-compatible API) to a set of tools the model can call: web search, code execution, browser automation via Playwright, file reading and writing, and more. The agent loop involves the LLM deciding which tool to use next, executing that tool, receiving its output, and repeating until the task is complete. A multi-agent mode called "flow" allows multiple specialized agents (such as a DataAnalysis agent) to collaborate on complex tasks. You would use OpenManus to automate complex research, data analysis, or software development tasks that would otherwise require human decision-making at each step. The tech stack is Python 3.12 with a TOML configuration file, Playwright for browser automation, and any LLM provider you supply API keys for. It installs via pip or the faster uv package manager.