CADRE

CADRE is a legacy family business, and like most of them it ran on a patchwork: spreadsheets, inboxes, manual approvals, and a great deal of knowledge that existed only in people's heads. The obvious half of the brief was to give that work a home.

The less obvious half came out of the audit. CADRE already had AI. What they did not have was any way for it to touch real work. Every task became the same round trip: download the files, upload them into a chat, explain the project from scratch, get an output, then file it back by hand. The assistant was capable and completely disconnected.

So Cadre Hub was built with two surfaces over one backend. Work is the browser workspace: boards, items, groups, typed fields, views, native docs with version history, dashboards built from real boards, files, sharing, activity and automations. MCP is an authenticated endpoint that lets Claude or ChatGPT operate that same workspace, and the business's Microsoft 365, as whoever is signed in.

The decision we would defend hardest is that Cadre Hub runs no model of its own. The client brings the reasoning; the platform supplies identity, permissions, live data, retrieval, tools, approvals and audit. That keeps CADRE free to use whichever assistant they prefer, keeps model costs off their platform bill entirely, and means the security boundary sits in one place rather than being reimplemented next to a model.

Retrieval is deliberately unfashionable. There is no vector database and no embedding pipeline. SharePoint discovery runs on a PostgreSQL lexical index over file metadata and the first slice of extracted text, combined with Graph search and deterministic ranking by project, document family, revision and date. Every candidate that could leak a filename, path or even the existence of a file is revalidated against live Graph for that user before it is returned, and an indeterminate result fails closed.

Browser and MCP paths converge on the same server actions and the same access checks, so an edit made by an assistant is not a second, weaker implementation of the product. Scope was held just as firmly: Work does not clone Microsoft's mail, calendar or SharePoint interfaces, and the forms module was archived rather than shipped half-finished.