What the integration covers
- Per-channel index: a distributor portal and a public storefront can hold different assortments, and each gets its own index.
- Kept current by events: catalog changes update the index automatically, there is no scheduled rebuild to wait on.
- Attribute filtering: narrow results by category, price range or any indexed attribute.
- Type-ahead suggestions: results appear as the buyer types, not after they hit enter.
- Hybrid mode: with semantic search enabled, the same index combines classic phrase matching with meaning-based matching.
How fast it is in practice
Search targets under 200 milliseconds at the 95th percentile, and the platform demo runs on a catalog of over 120,000 SKUs, so the numbers are not measured on a toy dataset. On a large B2B catalog with thousands of near-identical SKU variants, that is the difference between a search box people use and one they give up on.
The index is not a source of truth: the catalog lives in PostgreSQL, and the event that updates the index travels through the same Redis queue that handles webhooks and email. If the whole index had to be rebuilt from scratch, the source data is still in the database; nothing gets lost.