Causewaybay AI

Causewaybay Cafe puts Muse behind the counter of a small neighborhood cafe. Small cafes can't stop service to learn complex POS software, and guests often don't know the menu or the language. I built a cafe assistant that runs on the owner's Mac. Guests scan a table QR and just talk: "we are three, pick meals and drinks for us." Muse Spark reads the real menu, availability, basket and recent orders, then calls backend tools like order_now, checkout, order_cancel and staff_call. The model never writes to the database directly. The Rust backend checks role, item IDs, quantities and prices before committing to SQLite. "What's my order?" is answered from the database, not from the chat. Staff see the same order live through server events, prepare and serve it, and the owner sees tables, orders and hourly/daily token usage with hard limits. Muse Image makes an optional picture of the order, cached and rate-limited. Meta vision turns a menu photo or floor plan photo into an editable draft the owner approves. Stack: Rust (axum, sqlx/SQLite, HTTP/3 via quinn), Leptos WASM frontend, 7 UI languages. Business data stays local; inference uses Meta's cloud API. It's a working single-shop prototype: demo wallet credits, no real card payments yet.