Find the food you mean.

Local eligibility comes first

The pilot is designed around independent restaurants and small local groups. Major chains and franchises are removed before the ranking model considers taste, distance, meal occasion, or novelty.

Personal without requiring an account

You can browse as a guest. Likes, passes, saves, and other choices teach the taste model over time, including what works for different meals. The current release keeps that profile tied to the browser. Public account sign-in remains disabled; any future account system will require a cryptographically verified authentication gateway.

Safety is evidence, not a promise

The model evaluates allergen and dietary evidence at the dish level where possible. One conflicting dish does not automatically erase an otherwise useful restaurant, while shared-kitchen and venue-wide uncertainty stays visible. Users who need a stricter view can choose whole-place screening.

No recommendation replaces a direct conversation with the restaurant about ingredients, preparation, and cross-contact.

What is real today

This is a San Francisco MVP with a limited, reviewed catalog and some fictional pilot content. The product includes meal-aware discovery, taste learning, safety settings, location-based distance, saved places, and data export and deletion. Catalog depth and evidence coverage are still being expanded.