← Back

Security and the rep's mailbox

What Opptain reads, what it keeps, and what leaves the machine. Written 4 October 2026 for a security review; where the record lives added 5 October.

This describes a prototype. Everything above the last section is what the software does today. The last section is what does not exist yet, and a security team should treat those items as open.

Where the record lives

It depends on how the rep reaches Opptain. On Claude Desktop the record is a file on the rep's own machine, and everything below about "the rep's own machine" is literally that. In ChatGPT, or Claude on the web, an assistant cannot reach a laptop, so the record is held on Opptain's server (Google Cloud, United States): one record per rep, opened only with that rep's key, which the server holds encrypted under the rep's own sign-in token. There, read "the rep's own file" below as that record. Either way, mail is read only through the rep's own connection, and the server never holds a mail, CRM or call-recorder credential.

What it reads, and whose access it uses

Opptain has no mailbox of its own and no service account. It reads mail through the rep's own mail connection, under the rep's own credentials, on the rep's own machine, and only while the rep has it open. There is no central inbox, no shared credential and no background job: nothing is read when the rep is not there.

It searches, rather than syncing. A sweep runs a search against one account's correspondence — for example, threads with that customer since a given date — and reads what comes back. It does not copy the mailbox.

What it keeps

Two things are written, both in the rep's own file on their own machine:

What is storedContents
A record of the sweepWhen it ran, which account, the time window searched, the search terms used, how many threads came back, which mailbox. No message content.
One row per thread it usedSubject, time of the last message, how many messages, and a short note of what the thread established about the deal — which can quote what the customer wrote, with who wrote it.

Message bodies, attachments and recipient lists are not stored. The schema carries the rule in the file itself: the raw correspondence never leaves the machine; only the note does.

A rep can remove any of it — a thread, a contact, an account — with a reason recorded, so the deletion itself is answerable. Nothing is removed silently.

What leaves the machine

Three destinations, and mail is in none of them:

Put plainly: a manager cannot see a rep's mail, the notes taken from it, or which threads exist. They see what the rep sent them.

What a rep sees about their own mail

The screens say when a sweep found nothing — that absence is often the finding, as in the deal where no correspondence existed at all, which is how the product knew the buyer's review had not started. The rep is told what was searched and when, so the search is repeatable and arguable.

What does not exist yet, and should before any pilot

None of the following is built:

The honest summary

Architecturally this is the gentlest version of the capability: the rep's own access, no central store, notes rather than copies, and no path from a manager to a rep's correspondence. What is missing is the administrative layer a company needs in order to prove any of that — controls, retention, audit and the paperwork. That is build work, not a redesign.