Superapp is a chat that does the work: a claim is created, evidenced, reviewed, questioned, approved and tracked entirely in conversation — against real systems, with the human always making the consequential call.
Draft a claim, upload the receipt, review the OCR extraction, submit, approve, return, recall — the whole lifecycle runs in chat against a real backend with a full audit trail.
Lists render as typed tables, budgets as dashboards, decisions as confirmation cards — validated, rendered natively, never a wall of text.
Buttons like “Next page” execute the exact backend call they were built from — no model in the loop. Anything consequential still stops for a human confirmation.
The assistant can read what you may read and do what you may do — real permissions enforced by the backend, human confirmation on consequential actions, and an audit trail that remembers who did what and why.
Teams, Outlook and Calendar ride the same conversation — notifications land in chat, and replies go back out through Microsoft Graph.
Runs on self-hosted language models. Your prompts, documents and numbers stay on your infrastructure.
Your first five minutes: sign in, find your way around the workspace, chat with the self-hosted model, and switch on the Expense tools.
From a sentence to a submitted claim: draft it in plain language, attach the bill from your phone, review what the OCR read, and submit — with your explicit say-so.
Three kinds of refusal, all honest: a claim that changed under you, a claim with missing fields, and an action the state machine forbids.
Attach a policy document and ask questions against it — the answers come from your file, with the numbers to prove it.
Where administration lives: user accounts and roles, workspace settings, and the tool-server connections that power the assistant.
The same expense flow on a phone: chats in the drawer, the claim drafted in conversation, the receipt captured with the camera.
The flagship, end to end: a scanned hotel bill becomes an approved claim — claimant and approver, each working in their own conversation, one shared audit trail.
An approver's morning in one conversation: the pending queue, a full-context decision, an evidence request, and a return with named fixes.
The safety net: take back a submitted claim with a reason, read the timeline, ask the budget question, and check the conversational inbox.
The chat renders real interactive UI: a typed table for your expense list, a Next-page button that executes directly against the backend with no model in the loop, row-level View actions, and a live budget dashboard.