Did this land?
Off Grid AI Mobile · private alpha

Never forget
anything again.

Leave it running and your phone catches your day, writes it down, and hands you what mattered. The listening, the transcription, the insights - all of it happens on the device in your hand. Nothing is ever sent anywhere, because there is no server to send it to.

See what it does

The current private alpha is full. We are not taking more people into this round.

Think about everything you forgot today. The name you blanked on. The one line from the meeting you needed two hours later. The idea you had walking to lunch, gone before you sat down. Your memory leaks all day and you have learned to live with it. This is the thing that catches it, and it runs entirely on your phone.

What it does

Turn it on and it listens. A chief of staff who sat in every room you were in, caught every word, and forgets nothing - working while you get on with your day. This is the first real step toward ambient intelligence: help that is just there, without you opening an app to ask for it.

It listens, and writes it down
Your day, transcribed as it happens - the meeting, the call, the hallway aside you would have lost. Whisper runs on the phone, in resumable chunks, so a three-hour recording transcribes without eating your memory and picks up where it left off.
It finds what matters
A local model turns each recording into a summary, the key points, and the follow-ups you owe someone - not a wall of transcript you will never reread. Tell it what to watch for and give it the names and jargon it should get right.
You can ask it anything
Every transcript is indexed on-device and searchable. Open a chat about one recording and it stays in context for the whole conversation - "what did I promise her", "what were the numbers" - answered from the room you were actually in.
It knows your day
Reads your calendar, nudges you before a meeting starts, and files each recording against the thing you were actually in. The reminders work with no signal - they are local notifications, not a push from a server.
It keeps the speech, not the silence
An on-device voice-activity model gates the mic frame by frame, so silence never becomes a file and never gets transcribed. Compress a recording once it is analysed, or strip it to speech only and still restore the full timeline later.
You are the only one with a copy
No account, no upload, no transcript sitting in someone's bucket. The recorder is off until you turn it on, the app tells you when it is running, and you can delete any of it outright.

The part no one else will give you

Every bit of this happens on your device. The listening, the transcription, the people, the insights, all of it. Nothing is ever sent off your phone, not to us, not to anyone.

That is not a policy you have to take on faith - it is how it is built. Whisper runs in your phone’s memory, the analysis runs on a model you downloaded, and the search index is a database on your disk. Put the phone in airplane mode for a day and everything above still works.

Everyone else records your life onto someone else’s server. This never leaves your hand.

Sync is being built right behind it, so the same memory follows you across your devices, starting with your workspaces, projects, chats, and settings - over your own network, with no relay in the middle.

The honest part

Doing all of this on-device, privately, in real time is genuinely hard. Continuous capture plus transcription plus a local model pushes against real limits on battery and heat, and we are still finding the edges of it on real phones.

We would rather get it right with a group who run it every day than ship something rough to everyone. The current private alpha cohort is full. They run it daily, tell us what works and what does not, and shape what this becomes.