An inbox that reads itself before you get to it.

The problem

Somewhere in today's inbox is an email that matters — a deadline buried in an attached PDF, a customer document you've been waiting on, a request that needs a same-day answer. It's sitting between forty emails that don't matter, and someone on your team has to open, read, and triage all of it. When they're busy — which is always — the important one waits. Sometimes it waits too long, and "we missed the email" becomes an expensive sentence.

What we do

This is a build-to-order system, engineered for your inbox and your rules — not an off-the-shelf app you switch on. We assemble it from machinery we've already built and proven (see Receipts), then tune it to what actually lands in your inbox.

We build a watcher: an AI system connected to a business inbox that reads what arrives — the message and the attachments — figures out what each item is, extracts the pieces that matter (dates, deadlines, amounts, who it's from, what it's asking), and then acts on your rules. A deadline found in a fourteen-page document goes on the right calendar with a reminder. The person who needs to know gets notified — before anyone asked. Routine items get filed or routed. The genuinely unusual gets flagged to a human, with the AI's reading attached so triage takes seconds instead of minutes.

This isn't keyword-filter email rules. The system genuinely reads: a settlement document, a purchase agreement, an inspection report, an invoice — multi-page, messy, formatted by someone else. And it's built with the paranoia this class of system deserves: the watcher summarizes, schedules, and alerts, but consequential actions route through a human until you've watched it be right long enough to loosen the leash — and you define what "consequential" means.

How it works

1. Map the mail. What arrives in this inbox, what matters, what the perfect human assistant would do with each type.
2. Classification build. The watcher learns to recognize your document and message types on real (or realistic) samples.
3. Extraction rules. For each type: what to pull out — deadlines, parties, amounts — and where it goes.
4. Actions with a leash. Calendar entries, notifications, filing — automated; anything consequential requires human confirmation until you say otherwise.
5. Supervised weeks, then trust. You watch its decisions with full logs before it runs ahead of you.

What we deliver

• A watcher on the inbox you choose, reading messages and attachments as they arrive
• Automatic extraction of the facts that matter, defined by you
• Calendar entries, reminders, and notifications that happen before anyone asks
• A full decision log — every email, what the AI concluded, what it did

Receipts