The cultural intelligence chains buy, for the restaurants that can’t.

Chains decide menus, promotions and openings with data teams. An independent owner decides on instinct, between service and supplier calls. The cost of guessing wrong is a dish nobody orders, a quiet Tuesday, or a flyer handed out in the wrong neighborhood. Generic AI doesn’t fix this: it gives every tapas bar the same advice.

It starts from the people who already love a restaurant — as seen in Qloo’s cross-domain taste graph — and reads what else they love: nearby places, music, brands, films, dishes, vibes, even where in the city they live. Then a Claude agent turns that into three dishes to test, an event, a ready-to-post campaign aimed at the right neighborhoods, a supplier shopping list and a 7-day launch plan.

Every playbook carries a crowd lift: Qloo compares the restaurants this crowd loves most with the city’s most popular ones, and Tastecraft measures how much more often the crowd’s favourite places serve the proposed dishes. The same score is shown for the same model’s answer without Qloo, side by side.

• No personal data. Tastecraft only uses Qloo’s aggregated, privacy-first taste signals. It never sees or stores information about individual guests.
• Groups, not people. Qloo scores describe what people who like a restaurant tend to like, in aggregate. Tastecraft never presents them as a prediction about one guest, and keeps Qloo’s facts apart from its own interpretation.
• The right restaurant first. Every match is labelled exact, close or best guess, and the owner can switch to another candidate before relying on the results.
• Every fact is checked. The server records every value Qloo returns in a run and verifies each cited fact in the playbook against it. Unverifiable claims are flagged, not hidden.
• The owner decides. Tastecraft proposes; nothing is ordered, posted or booked without the owner.
• Honest gaps. Missing or thin data is listed in the playbook’s caveats. Playlists are inspiration: playing music in public still requires the usual performance licences.
• Qloo data stays private. Responses are cached server-side only and never published in the code repository.

• Who pays: independent restaurants and small groups (2–10 locations). A monthly subscription per location, with a free first playbook.
• Why now: agents can finally do the analyst’s job end to end, and Qloo makes the taste data available through an API.
• Distribution: restaurant software marketplaces (POS and reservations), food distributors who benefit from the shopping list turning into orders, and Supplora — the founder’s AI purchasing tool, used by a family-run restaurant in Spain.
• Expansion: connect sales data to close the loop (did the new dish sell?), multi-location playbooks for small groups, and concert or game-day playbooks around live events.

Antonio González Fernández, Business Administration and Data Science student at CUNEF (Madrid). He built Supplora, an AI tool that reads supplier delivery notes for a family-run restaurant and loads them into its management system — 715 notes processed and €1,876 of supplier price increases flagged so far. Watching that restaurant decide its specials on gut feel is where Tastecraft started.