Senses & Inbound Command
One brain that perceives the whole business.
Every signal that matters — inboxes, calendars, revenue, code, markets, domains, cash in the bank — flows into a single mind that watches all of it at once. But watching is all it can do on its own: it reads your email and drafts the replies, yet it cannot press send on anything without a human's yes.
Plugging in a new signal takes about 20 lines
Each "sense" is a tiny connector: read one outside service, report back what it saw. The platform handles everything else — it checks each sense is healthy, quietly retires the ones that stop mattering, and routes what they see: urgent items go into your daily brief, background facts go into the mind's long-term memory.
📬Email triage & follow-up radar
It reads your email and files the bills; junk is set aside where you can always retrieve it. It tracks who you owe a reply and which conversations went cold — with responses pre-drafted, never sent.
📅Meeting assist
A briefing before every call on your calendar, and afterward the transcript becomes notes, decisions made, and who's doing what.
📰Market & regulatory watch
New federal rules and market moves boiled down to what actually affects you — duplicates removed, noise filtered, delivered with your morning digest.
💵Cash runway
How many months of cash you have left, calculated from real bank balances and real spending. It can only look, never move money — and if a number is missing, it says so instead of making one up.
🌐Domain expiry
Watches your web-address registrations themselves, not just the certificates — an expired certificate is an embarrassment, but a lapsed domain is gone forever. It also sounds the alarm if someone else registers your brand name.
🔧Code review & ops briefs
Every recent code change reviewed for risk across all your systems, plus daily and weekly check-ins that end in one plain-English brief: what's done, and what needs your yes.
Privacy enforced in code, not in a policy binder
Seeing everything is powerful, so the walls are built into the machinery. Patient health information never reaches the business mind: an automatic screen inspects every incoming signal, and anything carrying patient data is turned away at the door. The clinical world and the business brain are kept physically separate.
- Consent comes first: the mind may build a picture of people who agreed to it — never of bystanders
- Every sense can only look — no sense has the ability to act, by construction
- Anything that would touch the outside world leaves the watching layer and enters the approval pipeline: rulebook, conscience check, then a written proposal for a human
- Feeds that stop being useful are automatically retired, so the AI can watch broadly without drowning in noise
By the numbers
Give it orders by email — behind eight locks
You can send the mind a task by email or chat message. That's also the easiest door for an attacker to knock on, so it was built backwards: this channel can never execute anything. Its only power is to write up a proposal for a human to approve — and it ships switched off until a human deliberately turns it on.
- Off by default — the channel does nothing until the owner deliberately switches it on, and a kill switch shuts it off instantly
- It listens only to one dedicated channel, never the open inbox
- A strict list of approved senders, verified cryptographically — not by the "From" line, which anyone can fake
- A rotating secret passphrase checked in a way that resists tampering
- Message contents are treated as untrusted — if a message tries to smuggle in instructions for the AI, it's refused and flagged as an attack
- Fixed rules — never the AI itself — decide how dangerous a request is; anything unrecognized is treated as maximum-risk
- Money, patient data, passwords, and anything irreversible are marked human-only — no message can ever authorize them
- If the conscience says block, the request dies on the spot — and every event is written to a permanent log
Design stance: shipping switched off isn't hesitation — it's the feature. Every high-risk capability arrives inert, proves its defenses under testing, and gains power only when a human decides it has earned it.