RAG / retrieval

In one line

Feed the model your own docs so it answers from facts, not guesses.

The flow

1. Load docs.
2. Split into chunks.
3. Embed with init_embeddings.
4. Store vectors (pgvector, Chroma).
5. Retrieve closest chunks and pass to the model.

Agent style: wrap retrieval as a tool so the agent searches only when needed. More flexible than a fixed chain.

Links

• Tools (retriever-as-tool)

My notes