Pelorus Query shapes RAG context from live demand.
Pelorus Query turns live RAG traffic into managed, source-grounded Extracts. It keeps chunk retrieval as the fallback, then lets repeated questions become visible, reusable, and curatable over time.
Chunking is an upfront bet
Every chunking strategy guesses how future questions will cut across the corpus. Smaller chunks, parent chunks, hierarchical retrieval, and graph-style structure can reduce the pain, but the retrieval surface is still mostly shaped before real demand arrives.
Pelorus adds a demand-shaped layer
Pelorus keeps chunk retrieval, then automates a second layer that is shaped by observed usage. Recurring needs become source-grounded Extracts, related queries become clusters, and the Admin views show coverage and gaps as demand changes.
Core Concepts
A query is the message your user or application sends to Pelorus. Pelorus keeps a durable representation of that query as an observed user need: something it can track, match, cluster, and use to create Extracts. An Extract is a source-grounded information packet built from your corpus to answer that need. Pelorus adds this managed Extract layer above the document chunks your RAG system already uses: specific queries get Query Extracts, and related queries can be grouped into Cluster Extracts that cover a broader area of demand.
Broader Extracts for groups of related queries.
Reusable Extracts for specific observed needs.
The indexed passages from your source documents that remain the evidence floor and fallback.