# DeepSearch Agents

A tutorial project that builds a multi-agent research assistant: a main agent delegates a task to web search, database and private knowledge base sub-agents, then merges their findings into a Markdown or PDF report streamed over WebSocket.

deepsearch-agents is a full-stack, runnable teaching project for a deep-research style assistant built with DeepAgents, LangGraph, RAGFlow, Tavily and FastAPI. A main agent analyzes the user's task, decides which sources it needs (public web search, a structured MySQL database, a private RAGFlow knowledge base, or files the user uploaded), dispatches work to expert sub-agents for each source, and merges their findings into a final answer that can be exported as Markdown or converted to PDF.

Each task runs in an isolated session directory identified by a thread ID, and a monitor pushes tool calls, sub-agent calls, progress and results to a React frontend over WebSocket in real time so the process stays observable rather than opaque. The repository is paired with a free, chapter-by-chapter written course and matching Git branches that walk through building the same system from scratch.

Website: https://didilili.github.io/ai-agents-from-zero/#/%E5%AE%9E%E6%88%98%E9%A1%B9%E7%9B%AE-%E6%B7%B1%E5%BA%A6%E7%A0%94%E6%90%9C/0-%E5%89%8D%E8%A8%80