Atlas
A local RAG assistant that searches personal documents and answers questions without sending them to a hosted model.
Problem
I wanted to ask questions across the documents I keep—bills, medical records, receipts, and more—without searching through them by hand or sending sensitive files to an external AI service.
Architecture
The full pipeline runs locally: documents are ingested, chunked, embedded, and stored in Qdrant. A question retrieves relevant document chunks and passes them to a local model for an answer. If the documents don’t contain enough to answer, Atlas says so. For questions unrelated to personal documents, it searches the web with SearXNG and brings those results into the conversation.
What I Built
Outcome
I can ask something concrete, like when I last changed the oil in my 4Runner, and Atlas retrieves the relevant maintenance record and answers from it. The documents and models stay local.