An AI assistant that can plan anything — and touch nothing — until you approve. It drafts. You decide. The box does exactly what you signed off on, and writes every step to a record no one can quietly edit. Not even us.
Most AI agents ask you to trust them with the keys. Ours is built so it never holds the keys at all. Three steps, every single time:
Whatever you ask for, the AI produces an inert, readable plan — plain steps you can inspect. The drafting engine is physically incapable of touching your files, your accounts, or the internet. It can only write.
Nothing proceeds without your cryptographic approval — one tap that only your device can produce. No approval, no action. There is no override, no "just this once," no fine print.
A separate executor — deliberately boring, with no creativity at all — runs your approved plan step for step in a sealed sandbox, then writes every action to a tamper-evident ledger you can audit forever.
Every promise below is enforced by how the software is physically constructed, and verified by tests that run on every release. If a promise fails, the product doesn't ship.
The creative AI is sealed off from your files, your network, and your accounts. It drafts plans; it cannot execute even one step of them.
Every plan is shown to you in plain language before anything happens. No background actions, no silent retries, no surprises.
Actions run only after approval that's cryptographically yours. Lose the signature, and the box simply stays shut.
The executor is deterministic — it does exactly what the approved plan says, the same way every time. No improvisation between your yes and the result.
Every event is written to an append-only, tamper-evident ledger. Anyone changing history breaks the seal — visibly.
"We didn't teach an AI to behave.
We built a box where misbehaving is impossible."
We're onboarding a small group of early users who want AI automation without the leap of faith. On your machine, under your signature, on the record.