Structured Search
Structured Search turns a Schema / AI Listing crawl into a filterable catalog your agents can search in real time (venues, cars, hotels, real estate, products, etc.). Users never see vendor names in the product UI — it is always labeled Structured Search.Why not only the Knowledge Base?
You can use both on the same agent. The search gate decides KB, Structured Search, or both per turn.
End-to-end workflow
1
Create a Schema / AI Listing crawl
Go to Crawler Jobs → New job → choose Schema / AI Listing.Paste one listing/index URL only — the page that shows all items (not a detail page; multiple source URLs are not supported).Example:
https://www.example.com/venues (or /cars, /properties, /products).Set max items, and start the job. Watch Live logs while discovery → schema design → extract runs.2
Review extractions items
Open the job page. Check Structured, Schema, and Failed tabs.Spot-check pricing, location, capacity, and gallery images before indexing.
3
Index into Structured Search
On the job page, open the Structured Search panel → Index into Structured Search.
4
Attach to agents
Click Add to agents and select one or more agents.Optionally edit the Agent tool description so the model knows when to search this catalog.
5
Chat & verify
Ask natural filter questions. In debug you should see Structured Search queries, filters, hit names, and injected context.Prompt the agent to render carousels/cards with
images: [url1, url2, url3] from result galleries (up to 3 images per card).Extend / update
Use Extend / update crawl when the site adds or changes listings. If the job was already indexed, it reindexes automatically when the extend finishes.Limits
- Paid workspaces only for index / attach
- 10 MB indexed payload per workspace (sum of all indexed jobs)
- Deleting a crawl job drops its index and detaches agents
Related
- Crawler overview — markdown crawl + schema crawl
- KB and UI Engine — cards / carousels from retrieved data
- Adding data to KB — classic markdown import path