Natural-language product search → ranked candidates + reasoning.
v1 reuses Phase 3-B’s tsvector engine. The query field is
treated as a plainto_tsquery argument (so multi-word queries
are AND’d against the indexed text). Match attribution
(match_reasons) is derived via simple substring detection
against the product’s name / description / tags /
vendor / product_type.
Defaults are tuned for an LLM context: top 5 candidates so the follow-up prompt stays compact. Caller can override up to 20.
Documentation Index
Fetch the complete documentation index at: https://docs.stella-commerce.com/llms.txt
Use this file to discover all available pages before exploring further.
Agent → Stella: natural-language product search.
query is a free-form natural-language request like "warm hat
for hiking under NPR 2000". Phase 6-C1 reuses the keyword +
trigram engine from Phase 3-B; semantic search lands later.
Successful Response
Response shape for POST /api/agent/intent/find.
Overall search-strategy summary (e.g. 'Matched against tsvector with cover-density rank; 5 candidates above noise floor'). Lets the agent surface trust signals to the customer without parsing per-candidate match_reasons.