Seven ways it watches your back.

It doesn't just guard agreements. It catches manipulation.

When an email claims terms you never agreed — payment terms, scope, deadlines — it's checked against everything in your history within seconds, flagged with the sources, the financial exposure, and a correction drafted in your voice.

It watches for the quieter plays too: gaslighting(“we never agreed to that” — when you did, in writing), responsibility transfer(“as you were supposed to handle…” — when they were), deadlines quietly sliding, commitments talked down after the fact, and a contact's tone shifting from partner to adversary. Each alert fires only when the language pattern AND your recorded history disagree — warnings with receipts, never paranoia.

While any risk flag is active, the AI cannot reply on that thread. Enforced in code, not in settings.

Every promise, remembered — both directions

“I'll send the deck by Friday.” “We'll confirm the PO by the 15th.” Promises made in email live nowhere — until they cost you trust. Every commitment is extracted with its owner and deadline, tracked until resolved, and surfaced before it goes overdue.

Yours keep you honest. Theirs give you leverage — politely, with the date and the quote, when you follow up. Over time it builds a reliability profile per contact: who delivers, who delivers late, and who never sends what they promised — so you calibrate before you depend on someone.

Ask anything. Get the answer with receipts.

Every fact, preference, and agreement is stored with its source, date, and a confidence score — searchable in plain language. “What did we agree with Acme about scope?” returns the answer and the emails it came from. You can also just tell it things — “draft a follow-up for Marcus”, “what do I owe Alex?” — and it answers from your data, never from imagination.

The memory gets sharper, not just bigger: the same fact confirmed in five emails becomes one strong memory, not five copies. When new information contradicts what's recorded — a number, a date, a story — it's caught as a contradiction, not silently overwritten. And when facts legitimately change, the old version is kept with its history: who changed what, and when.

After three months it knows things you've forgotten. Software can be copied — your history can't.

Signals no single view can see

One layer watches health, another watches risk, another watches the calendar. The insights engine reads them all at once and catches what only the combination reveals: a deal entering its close window (engagement up, decision-maker copied in, late-stage language), a relationship quietly entering pre-churn, multiple danger signs aligning on one account, or a long-silent contact suddenly re-engaging.

It learns your world's rhythms too — and calibrates itself against its own track record, so warnings get sharper the longer it runs.

Every insight lands in your brief with the signals that produced it — act on it in one click.