# LangManus LangManus is an open-source AI automation framework that combines language models with specialized tools to perform web search, crawling, and Python code execution. It implements a hierarchical multi-agent system using [[LangChain]] and [[LangGraph]]. MIT license. Built by a group of former colleagues as a community/research project. Motto: "Come From Open Source, Back to Open Source." ## Architecture Seven specialized agents organized hierarchically: 1. **Coordinator**: entry point; handles initial interactions and routes tasks 2. **Planner**: analyzes tasks and creates execution strategies 3. **Supervisor**: oversees and manages execution of other agents 4. **Researcher**: gathers and analyzes information via web search 5. **Coder**: handles Python/Bash code generation and execution 6. **Browser**: web browsing and information retrieval (uses Playwright) 7. **Reporter**: generates reports and summaries of workflow results ## Three-Tier LLM System - **REASONING_MODEL**: complex reasoning and decision-making - **BASIC_MODEL**: simpler text-based tasks - **VL_MODEL**: vision-language tasks involving images Each tier has independent API key and base URL configuration. Supports open-source models (e.g., Qwen) and any OpenAI-compatible API. ## Key Features - Multi-agent orchestration with hierarchical delegation (supervisor pattern) - Web search via Tavily API - Neural search and content extraction via Jina - Built-in Python REPL and code execution environment - Browser automation via Playwright - FastAPI-based API server with SSE streaming - Agent behavior defined in markdown prompt files (easy to customize) - Web UI available separately ## Technical Details - Python 3.12+, dependency management with `uv` - Business logic embedded in prompt markdown files (`src/prompts/`) ## References - https://github.com/langmanus/langmanus - https://github.com/langmanus/langmanus-web ## Related - [[LangChain]] - [[LangGraph]] - [[AI Subagents]]