Work

No accounts and no server round-trip — licences verify with HMAC locally, so it works with the wifi off.

It learns your voice from what you've already published, so the output reads like you wrote it rather than like it came out of a prompt box.

Built to empty, not to fill. Ticking something off removes it from the panel — the defence against becoming another graveyard of saved links.

No vision API and no OCR. Local Whisper only, so nothing leaves the machine and it costs nothing to run.

Grown by digesting references rather than copying them: every lesson is written as a checkable rule with a reason, not a description.

The identification gate was verified at 100% across 25 real photos before any of the collection mechanics were built.

The rep state machine rejects partial reps — verified against real footage, because a counter that flatters you is worse than none.

Every icon and value was extracted from real frames of the original — nothing drawn from memory.

The camera was tuned against measurements taken from Screen Studio's own demo footage — dead-still holds, spring ease-out pans peaking at 18%, 0.42s zoom-in against 0.84s out.

The school pays; parents never pay for the core and drivers never pay at all. Pricing designed so the safety features can't sit behind a paywall.

Driven by hooks and keystroke injection, so it works with tools that were never built to expose an API.

A drop-shelf and timer that live in dead pixels you already own.

Closed-set keyword spotting rather than full speech recognition — a much smaller, faster problem that runs on-device.