yongtenglei/ragflow

RAGFlow helps you build AI assistants that actually know your business. Instead of a generic chatbot that makes things up, it connects an AI model to your own documents, PDFs, Word files, spreadsheets, presentations, scanned images, web pages, so it can answer questions grounded in real data you control. Think of it as giving a language model a curated library of your materials to reference before it responds. The way it works is straightforward conceptually. You upload your documents, and the system reads through them carefully, breaking them into intelligent chunks using template-based rules (not just blindly splitting text every N words). When someone asks a question, it searches those chunks for the most relevant pieces, shows you exactly which passages it found, and passes them to an AI model to generate an answer with traceable citations. You can see the chunks, adjust them, and verify where answers came from. It supports connecting to various AI model providers (like OpenAI's GPT series or Gemini), and can sync data from sources like Notion, Google Drive, Confluence, S3, and Discord. This is built for teams who need reliable, document-grounded AI, legal firms searching contracts, support teams answering from knowledge bases, or enterprises with large internal document collections. A founder could use it to build a product documentation assistant, a PM could create a tool that answers questions from specs and meeting notes. The agent workflow capabilities also let you chain steps together, like running code or following a multi-step process. The project runs via Docker (a packaging tool that bundles everything needed to run the software), and requires a reasonably capable machine, at least 4 CPU cores, 16 GB RAM, and 50 GB of disk space. Notably, it only ships x86 Docker images currently, so ARM64 users (like Apple Silicon Macs) need to build their own. It uses Elasticsearch for storage by default but can switch to an alternative engine called Infinity.