What Proxara Connect is

Proxara Connect lets a firm's people ask Claude for work that lives inside the firm's own systems, without handing client identities to the model. It links Claude to Microsoft 365: Outlook mail and calendar, Teams messages, OneDrive and SharePoint files, and tasks. Support for other assistants is not available yet.

Nothing installs: no agent on anyone's laptop, no certificate, and no device management. The Microsoft side is one administrator consent, granted once for the whole firm, after which each employee signs in with their own Microsoft account from inside Claude.

What it does

An employee asks Claude to get them ready for a meeting. Connect prepares a brief for one meeting on that person's calendar, from an hour ago to two weeks ahead, drawing on the mail, chats, files, and tasks the firm already holds. If no single meeting matches, it says so rather than guessing.

Retrieving and acting are two separate consents. The one granted to start reads; drafting a reply or creating a task in Microsoft 365 runs under a consent the firm grants separately, and the model proposes that work rather than performing it. How actions work covers the proposal, the approval, and what commits it.

How protection works

Before anything reaches Claude, Proxara's policy engine reads the request with the firm's own governance rules and enterprise context. That evaluation runs in Proxara's environment rather than on the employee's device, and part of it is a classification step on a model Proxara operates.

The engine replaces the people, organizations, accounts, and other protected references with stand-ins, then passes the transformed material to the model. A stand-in reads as [Person_0CE2473EA47B]. It is computed under a key held for that firm, from inputs that never include a name, and the same client receives an unrelated stand-in next time, so nothing accumulates into a profile on the model's side.

The model is never shown who sent a message or who is on an invitation. After the substitution Proxara re-reads what the model would receive, and if any real name or address from a message survived, the whole request is stopped rather than sent.

The employee sees the real result in a Proxara view inside the chat. That view is the only place a real value is restored, it opens only for the employee whose work produced it, and a stand-in copied into another conversation resolves to nothing.

What is kept

Connect keeps no second copy of the mailbox, calendar, or file store. Each brief reads what it needs at that moment, and the working copy, the encrypted private view, and the stand-in mapping all expire an hour later. Once the hour is up, every route refuses them.

What lasts is the record. Every request, its outcome, and each time an employee opens the private view land on the firm's signed record, which carries counts, categories, sources, and timestamps rather than client content or client names.

The posture, stated plainly: Proxara enforces the firm's approved AI-use policy and narrows disclosure. It does not by itself create legal compliance.

Alongside Endpoint Protection

Connect governs what AI reads out of the firm's systems. Endpoint Protection is the device product, and it governs what a person types into an AI tool. Both evaluate the same signed rule library, and the console keeps them apart: a firm running only Connect is told plainly that no endpoint coverage is inferred from connector evidence.

Connect or Endpoint sets out which to start with.

Where to go next