Tablekeeper by Culltron (Nightshift)

Tablekeeper is a calm reservation app for one restaurant book. A diner searches for a time and party size and sees only the tables that are really free. The table she picks stays selected while she signs up or signs in, so she never starts over. Bookings need no reference code: they live under My bookings, where she can move to another time or table, or cancel with a confirm step and a Keep reservation option. After every search, booking, move, and cancellation, three agent seats (coordinator, implementer, reviewer) each return a short note from a live model (gpt-5-mini on Azure OpenAI). The reservation engine stays the only authority: it validates every action once, prevents double bookings, and handles joined tables, retries, and restaurant-local times. If the model provider fails or runs out of credits, the app keeps working and clearly shows that the seats are simulated. Only the action name and its result are sent to the model. No diner details, references, or session tokens leave the service, and keys are read from the environment only. The service passes 68 local tests and is deployed on Azure Container Apps. It is a reference build for the Tablekeeper track, not a judged Band room run.