Building a shopping agent on your live catalog

Most shopping assistants answer from a saved crawl of your web pages, not your live catalog. So they recommend sold-out products and quote last month's price at their best. And the assistant only surfaces what actually exists and passes your rules when the retrieval is right, which a crawl can't guarantee.

Sanity Context connects your agent to your store's real data through one MCP endpoint. Hard filters (in stock, under budget, right size) run in the same query as keyword and semantic search, so the agent returns real products, filtered by your rules, ranked by relevance, from your live dataset. Update a product or a price and the agent knows immediately.

In this live demo, we'll build it: from a store with no agent to a working shopping assistant grounded in your live catalog.

What we'll cover:

• The 5 min setup: Connecting your agent to your content with Sanity Context, so it's grounded in your live catalog and only surfaces products that exist and pass every rule (stock, price, size)
• Combining exact filters (in stock, under budget) with search that understands what shoppers mean, so "comfy jacket" finds the right products even when that word isn't in your catalog. One search, no extra service to run
• Governing how the agent responds from Studio (its tone, what it can and can't recommend), with no code deploy
• Staging and previewing agent changes in Content Releases before customers see them
• Adding personalization and cart-aware answers for logged-in shoppers, plus the exact stack we used so you can rebuild it