Hire a team that reads everything and forgets nothing.

Classifies every email — pipeline signal, personal, vendor noise, internal ops — and extracts typed signals: new stakeholders, timelines, budgets, competitors, commitments made by either side, risks, and who's waiting on whom.

Every extracted signal carries a verbatim quote from the email it came from. Proposes new contacts and close-date changes into your approval queue — it never writes them itself.

One screen: the single objective for the meeting, the deal's current state, what changed since last contact, and who's in the room — each attendee's stance, with a flag for anyone you've never met.

Exactly three talking points and at most two risks to probe. If a stance can't be backed by a verifiable quote, it says “no signal yet” instead of guessing.

Grades health as a category — green, yellow, red — backed by evidence, because a two-decimal health score is fake precision. Runs eight named risk checks and names THE single next action.

Fills gaps in your win plan with evidence-backed suggestions rendered under the field they target — and when the record contradicts what you wrote, it says “Sentinel questions this.” Your click is the approval.

Watches for milestones slipping, unstaffed weeks ahead, budget overrun risk, burn running ahead of plan, and clients gone quiet.

Proposes a concrete project task to defuse each risk — into the same approval queue as everything else.

Annotates the forecast in 120 words or fewer. Every figure it cites is pre-computed by the system and handed to it as a numbered fact — the narrative must cite those facts, and citing one that doesn't exist fails the run.

The published note discloses its run id, with every computed fact it cited one click away.

A headline on ARR versus pace, then at most three ranked items that need a human today — each with a button that takes you straight to the work — plus what the agents did overnight.

Hard-capped at 250 words. If an item can't produce its evidence, the item is dropped, not the standard.

Studies your real sent mail — tone, greeting, sign-off, sentence length — and drafts one clear ask per email. Anything it inferred rather than knew is flagged in an uncertainty note.

Drafts are permanently approval-first. Once you approve, the email sends from your own mailbox — not a noreply address.

Ask across live sales and delivery data: accounts, contacts, deals, pipeline and forecast summaries, projects, capacity, tasks.

Its only write path is proposing an action into your approval queue — the Assistant follows the same constitution as every other agent.