Plain-language questions, cited answers computed from your own live data — never made up. Ask in the manager portal, over WhatsApp, through the API, or from AI tools like Claude Desktop connected straight to your restaurant.
"Best seller in September", "labour % by outlet last month", "top 10 wasted items" — asked the way you'd ask a person, answered with numbers.
Every figure in every answer comes from an executed, logged query over your own live data. If it can't be computed, the answer is an honest "I couldn't compute that" — never a guess.
Answers draw only on governed reporting views — sales, labour, covers, inventory variance, loyalty — with business definitions baked in, so "net sales" always means net sales.
Generated queries are parsed, restricted to read-only views, scoped to the outlets the asker may see, and time-limited — the engine can look, never touch.
"Who last changed the taco recipe?" "When did Aina start planning the buka puasa event?" — answered from the platform's activity memory, citing the audit record.
Each answer carries the key figures, a compact chart, and a "how this was computed" disclosure — the exact query and views used, one tap away.
Any answer re-runs at finer grain — by outlet, by week, by daypart — so a headline number becomes a breakdown without retyping the question.
Message your restaurant's number from the pasar or the school run — same engine, same security, answers in the chat you already live in.
A clean ask endpoint returning structured answers, plus direct endpoints on the governed views — feed your own dashboards and spreadsheets from the same trusted layer.
Point Claude Desktop or any MCP-capable client at your restaurant's own MCP endpoint and query your business inside your own workflow — token-authenticated, permission-scoped.
Question, query, execution stats and answer recorded per asker and channel, with monthly AI budgets applied — full accountability for every answer given.
Guest-level personal data stays out of the answer layer by default — aggregates only, with named-guest questions permission-gated and logged.
Ask "what was our highest selling item in September?" — the engine writes a query against your governed sales views, validates it as read-only and scoped to your outlets, runs it, and answers with the figures, a mini chart, and the SQL collapsed underneath. Numbers you can defend, because you can see exactly where they came from.
"Who last changed the taco recipe — and did its cost move after?" The memory side finds the change, the actor and the timestamp from the audit trail; the data side queries recipe costs around that date. One composed answer, both sources cited. It's the question every owner asks and no reporting tool can answer.
Connect Claude Desktop to your restaurant's MCP endpoint with a device-bound token and ask "labour % by outlet last month" mid-workflow — same governed pipeline, same outlet scoping, every call logged. Nothing is reachable without a valid token, and answers never exceed what that person is allowed to see.
Be the first — join the waitlist and get onboarded first, on your own private server.