routing

Query routing automatically directs search to the optimal data source—vector database, web search, or direct LLM—based on query type. This eliminates manual selection and ensures product searches, policy inquiries, and troubleshooting queries reach the correct collections and search strategy.

Vector database routing searches internal knowledge collections (catalog, FAQ, troubleshooting) for structured answers; web search routing queries external sources for current information. The routing decision selects the best method based on query intent and available data.

Yes. Routing identifies when a query requires multiple collections—for example, searching both troubleshooting and FAQ collections when a customer asks about technical issues and return policies—and selects the appropriate collection strategy accordingly.

Routing enforces consistent decision logic across retry attempts, returning a stable routing decision with route, collections, strategy, and reasoning. This prevents query instability and ensures the same query type receives the same destination.

A routing decision returns the selected route (vectordb, web_search, or direct_llm), the target collection set, the search strategy, and reasoning for the choice. This metadata enables downstream processing to execute the routing plan correctly.

Routing initiates at the pipeline start to determine data source direction before any search or retrieval occurs. This early classification prevents wasted computation and directs all subsequent steps—search, collection selection, and strategy execution—to the correct target.