knowledge-query

Hybrid retrieval merges vector similarity with BM25 keyword matching using Reciprocal Rank Fusion (RRF) to locate factual information in a pgvector knowledge base. This combination ensures both semantic context and exact keyword relevance improve search accuracy.

You can search your knowledge base by querying with a command like "@knowledge <query>". Setting the answer parameter to true synthesizes a narrative response grounded in retrieved snippets and appends citations for source verification.

Yes, knowledge search supports targeted filtering. You can narrow retrieval results by specifying connection, space, unit, content type, topics, and date range to apply metadata-boosted filtering against the indexed knowledge base.

Using RAG with hybrid retrieval is the best way to fetch top-k chunks. It queries the pgvector knowledge base using RRF to combine vector and BM25 results, returning relevant chunks for question answering or targeted retrieval.

The search mechanism handles connection and space availability errors. If a specified connection or space is missing or unavailable, the system manages the error rather than failing silently during the knowledge base retrieval process.