Retrieval-Augmented Generation
The model answers with evidence from your files
Follow a local document as it becomes searchable, then follow a user question through retrieval, reranking, grounded generation, and citation checks. Select any process to inspect what goes in and what comes out.
Static, offline customer walkthrough
Embedding
qwen3-embedding:4b
Reranker
BAAI/bge-reranker-v2-m3
Vector store
LanceDB · local
Stage 1 · Prepare knowledge
Runs when local files are indexed or updated
5 processes
Stage 2 · Answer a question
Uses the prepared index for every prompt
7 processes · reranker on
How Stage 1 powers Stage 2
Your files build a searchable local index. They do not retrain the chat model.