RAG Chat · Ask your documents

RAG Chatask your documents · runs in your browser

Answer model checking GPU… GitHub

Library

empty

Add documents PDF, Markdown or text · drop or click

How it works

Chunk: text is split into ~900-character passages along paragraph and sentence boundaries.
Embed: each passage becomes a 384-d vector with bge-small-en-v1.5.
Retrieve: your question is matched by meaning (cosine) and keywords (BM25), merged with reciprocal rank fusion.
Generate: an LLM running on your GPU answers only from those passages and cites them.

Your files never leave this device.

Sources

ask something

The passages used for an answer show up here, with page numbers and how they matched.