Separate RAG database¶

Build a second, independent knowledge base — a one-off database you create and point a room at — alongside the ingester's continuously-updated one. This builds on Populate RAG, so do that first (it covers ingesting a corpus and querying it from a room).

1. Build a separate, static database¶

The haiku-ingester service maintains one continuously-updated database (rag/db/haiku.rag.lancedb). For a separate corpus you build once, ask the agent to create a standalone database — say a handbook from another directory of Markdown. The skill builds rag/db/handbook.lancedb as a one-off, independent of the ingester's database (and it refuses to overwrite the ingester's own database while that service is running).

2. Point a cloned room at it¶

Ask the agent to clone the custom room under a new id — say handbook — and wire that copy to the new database (rag_lancedb_stem: "handbook") instead of the ingester's. Open the new room and ask a question answerable from the handbook corpus: it answers from that database, while the custom room still uses the ingester's.

A room is not limited to one: list both under rag_databases and a single search covers them together. See Searching several databases at once.

Where next¶

Back to Next steps for the other optional walkthroughs.