Where your audio actually goes

The honest version of the pitch is this: IntelliKin only works if you let a microphone listen to your day. That is a real thing to ask of someone, and the only reasonable response to it is to explain, in detail, what happens next.

So here is the entire path a moment of audio takes, and where it stops existing.

The short answer

Raw audio is never stored. It is deleted twice, at two different points, by two different systems, and what remains is text you own and can delete yourself.

The long answer, step by step

Your watch captures and compresses on device. Audio is encoded to Opus right on the watch, which keeps files small enough to move without shipping anything raw across the network.
It uploads over an encrypted connection to our servers, and nowhere else.
Silence is discarded immediately. Voice-activity detection runs first. Files with no speech in them, which is most of them, are dropped before anything else looks at them.
Speech is grouped into conversations. Chunks get stitched back into runs of talking, split where there is a long gap. A conversation, not a stream of fragments.
It is transcribed by a speech-to-text provider under a contract that forbids using your data to train their models.
Your voice is separated from everyone else’s so we know which half of the conversation is your commitment and which half is something you merely heard.
Claude reads the transcript and pulls out the tasks, decisions, and details worth keeping.
The text is stored. The audio is deleted. That is the end of the recording’s life, usually within minutes of it being captured.

Steps 3 through 8 all happen on our servers rather than on your phone, which is a deliberate choice: it means we can improve the pipeline, or swap a provider that stops meeting our standards, without waiting on an app update to reach you.

Deleted twice, on purpose

There are two copies of a recording in the world at its peak, and both are scheduled to die.

Your phone holds its copy only until the server confirms it has the upload, then deletes it. The server holds its copy only until the pipeline finishes, then deletes it. Neither waits for a nightly cleanup job or a retention window to expire. Deletion is a step in the pipeline, not a policy we promise to remember to enforce.

If the pipeline fails, the audio is still deleted. We would rather lose a conversation than keep one.

About your voiceprint

To tell your commitments apart from things said near you, we build a voiceprint: a mathematical signature of your voice. It is a biometric identifier, and we treat it as one.

We create one for you only. We do not build voiceprints of your colleagues, your family, or anyone else who happens to be in the room.
It is stored securely and used for exactly one purpose: recognising you.
It is deleted when you delete your account.

Everyone else in a conversation is simply “not you”. That distinction is all the product needs, so it is all we keep.

What we do not do

No advertising. No selling data, to anyone, in any form. No using your conversations to train models, ours or anyone else’s. There is no version of the business where your day becomes someone else’s inventory. We would rather charge you money.

You can see all of it

Everything Kin has learned about you lives in one screen in the app, written in plain language, organised by person and project and topic. Not a settings page buried three levels down, and not an export you have to request by email.

You can read it, correct it, and delete any of it. Deleting a fact deletes it, with no tombstone and no shadow copy. And if you want the whole account gone, that works too, in one action, with no retention clause and no exit interview.

Why we are this specific about it

Most privacy pages are written to be defensible. We would rather this one be useful, because the actual question you are asking is not “is this legal” but “what would I find if I looked”.

This is what you would find. If something here changes, we will say so here first. The formal version lives on our privacy page, and if anything in either is unclear, we would genuinely like to hear about it: contact@intellikin.com.