Est. 2026 — General goods
Shopping for people
who read the spec sheet.
Describe what you need in your own words. The assistant reads the actual specification tables — and your own order history — rather than guessing from marketing copy.
Try: “laptop under £1200 for video editing”
New in
Browse all →
Under the hood
The assistant queries the database. It does not improvise.
01
Grounded in real rows
Product answers come from tool calls against the live catalogue — the same search function the storefront uses. Policy answers are retrieved from policy documents by vector similarity.
02
Isolated by Postgres
When it reads your orders it queries as you. Row-level security decides what is visible, so it cannot surface another customer’s order — the rows are not in the result set.
03
Cheap where it can be
Recommendations are a pure vector query with no model call at all. Search spends one small model call to turn your sentence into a filter.