How it works
How it stays private.
Stonekept lives across your iPhone, your Mac, and if you want it, Claude Desktop. Here is what happens to a recording at every step, shown in the real app. No marketing fog. No claims you can't verify.
It starts with one tap.
Tap your Apple Watch, click your Mac menu bar, or open the iPhone app. A Live Activity shows the mic is on the entire time it's on. No wake word, no background listening, and the mic stays off until you tap record. A session can be a thought or a whole afternoon.
Capture is the only step that ever touches a microphone.
Already have a recording? Bring it.
Drop in a Voice Memo, a handheld recorder file, a meeting export, or an old interview. We take m4a, mp3, wav, aiff, flac, and opus, and it picks up the same on-device path as a live session.
Your phone reads it.
The moment a session ends, your iPhone transcribes it on the iPhone itself. Apple's on-device speech model runs first, with a second on-device model as backup for a noisy room or accented speech. There is no upload step here, even on Pro.
Your voice never reaches our servers.
The parts worth remembering, lifted out.
Nobody listens back to a 47-minute file. Stonekept's on-device AI reads the transcript and lifts out what your brain would have kept: decisions, tasks, people, patterns, and ideas. Each becomes a structured memory you can search and link. The model downloads once, runs on your iPhone's own processor, and never calls home.
On-device AI. No cloud. No model API. The phone does the work.
Hand the heavy lifting to your Mac.
Extraction is the heavy part, and on iPhone it runs in the foreground while the app is open. Pair a Mac and Desktop Assist takes over: it runs the larger model, finishes hands-off, and spares your phone's battery. Your captures land in Memories either way.
The phone captures. The Mac does the heavy lifting.
Nothing lands in your Memories without your yes.
Every extracted memory lands in a queue first. Keep what matters, edit a fact, toss the rest. You see the source quote, how confident the model was, and the conversation it came from, on your iPhone or on the Mac.
The AI drafts. You decide.
Sync without seeing inside.
Add a second device and your memories follow you. Each one is locked on-device with a key only your devices hold. We move the locked package through our servers but can never open it. The keys never leave your hardware, so your memories stay yours alone, readable by you and no one else, not even us.
Encrypted in transit and at rest. End to end. Real.
Ask anything about your memories.
Open Chat and talk to your own life. Ask what you decided about the v2 launch, or what someone said on Tuesday. The model that answers runs on your iPhone or Mac. No sign-in, no cloud, no catch.
Same on-device AI that does extraction. Different prompt.
Bring your memory to Claude.
On Pro, you pick what Claude can read, and Stonekept hands it over only when you ask, only what you chose, only for that conversation. On your Mac it runs local-to-local, so nothing leaves the computer. Away from it, an encrypted broker relays the chosen memories and stores nothing. Bring your own Claude key; we never mark up tokens.
Scoped, read-only, revocable. Turn it off any time.
A note on model sizes
Different devices run different sized models.
Stonekept Extract comes in two sizes, Stonekept Small and Stonekept Large. The one your device runs depends on how much memory it has. The same things come out of either size (decisions, tasks, people, ideas); the larger model just writes longer verbatim quotes and picks up more of the detail.
Extraction runs entirely on-device. Pairing a Mac (Desktop Assist) is optional: it speeds things up and spares your phone's battery, and because the Mac can run Stonekept Large, you get higher-quality output even on a phone that only runs Stonekept Small.