AI can guide the guest. Staff still own the service.
TableSage avoids overconfident answers where availability, dietary safety, or staff judgment matters.
Menu-grounded only
The AI only recommends items present in the live restaurant menu. Invalid recommendations are discarded server-side and logged for owners.
No invented prices or dishes
Prices, ingredients, allergens, and modifiers come from approved menu data — not the model's guess.
Allergy review enforced
Allergy and severe-dietary questions surface 'confirm with staff' language and route to the staff queue automatically.
Availability respected
Unavailable items are never recommended.
Safety-sensitive language
Alcohol prompts, medical claims, religious dietary claims, and severe allergy language avoid overconfidence.
Tenant isolation
RLS scoped to each restaurant. Anonymous guest writes are constrained to session-level rows.
Server-only secrets
Provider API keys live in server functions only. Never in frontend code or repo history.
Demo isolation
Demo seed data lives in its own tenant — disconnected from any production restaurant.