Hybrid retrieval, typed
Keyword full-text, vector similarity, and structured filters combine through reciprocal-rank fusion. You declare the channels in TypeScript; the schema and SQL are generated per project at runtime.
samesake
Search built for shoppers who don't know the product name — screenshots, vague intent, budgets, occasion, and availability. You declare the catalog and retrieval in TypeScript; samesake compiles it into a Postgres-backed search layer you run inside your app.
What it does
Section titled “What it does”Hard filters stay hard
“Under 20,000” and “available now” compile to typed constraint predicates pushed into SQL — not soft semantic vibes. Every search carries an auditable constraint trace.
Image and intent together
Image similarity, text intent, price, freshness, and availability are separate, query-time-weighted signals. Tune visual vs intent vs price without reindexing.
Two containers in production
Postgres (pgvector + pg_trgm) and your app process. Bring your own embedding and generation models. No Redis, no Elasticsearch, no hosted vector DB.
New to this?
Build a search experience walks from an idea to a working search, in plain language.
Already know what you want?
Quickstart — collection to push to index to search in about fifteen minutes, no LLM required.
Have a store already?
Wire it up: MedusaJS, headless Shopify, or headless WooCommerce.
Want proof?
Eval from search snapshots builds a corpus from real catalogs and scores keyword vs hybrid.