Your files, your machine
How document chat works
Most document chat uses retrieval-augmented generation (RAG). Your files are split into chunks and turned into numbers that capture meaning (embeddings), stored on your computer. When you ask a question, the app finds the most relevant chunks and hands them to the model with your question. The model answers from those passages.
Tools that do it locally
AnythingLLM pairs with Ollama or LM Studio and organises documents into workspaces. Jan lets you upload files into Projects. Open WebUI offers a browser interface for a local engine. Our Windows guide walks through Ollama plus AnythingLLM step by step, with no graphics card required.
Where it falls short
Retrieval can miss the passage that matters, especially in tables, scanned PDFs and long documents. Questions that need the whole document at once ('what are all the deadlines in this contract?') are harder than finding one fact. Ask the tool to quote and cite the passage, and open it to check anything important.
Keep the whole chain local
Document chat is only private if every part is local: the model, the embedding model, and the storage. Some tools default to a cloud embedding or a cloud model. Check the settings for each, and do the offline test with a document question, not just a chat question.
Which documents would you never paste into a cloud chatbot, but would love to ask questions about?