Claude + open source research stack is getting stupidly powerful 7 repos can turn one model into a full autonomous research system 01 Firecrawl ▸ → turns messy websites into clean data the model can actually use 02 Browser Use ▸ → lets the agent click, scroll, search and interact with the web 03 Crawl4AI ▸ → crawls entire sites and structures the useful content 04 Qdrant ▸ → stores everything the agent finds so it can retrieve it later 05 LangGraph ▸ → controls the research loop and decides what happens next 06 Docling ▸ → converts PDFs, reports and documents into model-friendly data 07 Langfuse ▸ → shows exactly what the agent searched, used and generated the loop: → user asks a question → LangGraph builds the research plan → Browser Use searches the web → Firecrawl + Crawl4AI extract the useful pages → Docling handles PDFs and reports → Qdrant stores the findings → Claude connects the evidence → Langfuse records the entire run → final research report It becomes the brain coordinating a stack of tools that handle browsing, crawling, memory and documents.