
On your phone.
Hold the orange button and speak, or type. Notice answers out loud, in any of Apple's voices or your own Personal Voice. The transcript waits behind Face ID in the Memory tab.
Hold the button, say what is on your mind, let go. Notice hears you with on-device speech recognition, thinks with an open language model running on your phone's own chip, and answers out loud. No account, no server, no API key.
Needs an iPhone with 8 GB of memory (iPhone 15 Pro or newer, iPhone 16 or newer) and 2 to 5 GB free for the model.



Every word is heard, understood and answered on your iPhone. The watch and the wearable just carry your voice to it.

Hold the orange button and speak, or type. Notice answers out loud, in any of Apple's voices or your own Personal Voice. The transcript waits behind Face ID in the Memory tab.

Hold the dot, say it, let go. Your watch carries your voice to the phone, the phone thinks, and the watch speaks the answer back. Nothing else in the chain.
Watch app coming after the iPhone release
A small round face, one orange button, a loop for your wrist. It talks to your phone over Bluetooth and to nothing else. Open hardware, designed in the open.
Buy one or build one, below


One button. Press and hold to talk; let go and it answers through its own speaker.
Pairs once. A code on its face, the first time only; after that it finds your phone by itself.
Nothing inside to leak. No model, no key, no conversation lives on the device. It is ears, mouth and face; the phone is the brain.
The wearable will be sold two ways: assembled and ready to pair, or as the plans to print and build your own. Both are being finished now.

Printed shell, board, battery and cord, assembled and tested, with a wrist loop. Pair it to the app and talk.

The printable parts, the parts list and the illustrated step-by-step guide, for anyone with a 3D printer and an afternoon. You buy the board and the small parts yourself.

A mind you choose. Download an open model once and keep it. Qwen3 4B is the one we tested; Llama 3.2 3B and Gemma 3 4B are on the shelf, and you can add any MLX text model from Hugging Face by name. The download is the only time Notice uses the internet.
A memory that is yours. The conversation lives in one encrypted file on your phone, behind Face ID. Erase it with one tap. Never backed up to a cloud, never synced anywhere.
A voice you like. Any of Apple's voices, or your own Personal Voice.
An optional wearable. Notice also pairs over Bluetooth with the open-hardware Notice wearable: a small round face and one orange button, so you can talk without taking the phone out. Everything still runs on the phone.
Notice has no account, no analytics and no ads. Your speech becomes text on the phone, the text goes to a model running on the phone, and the answer is spoken by the phone. The only time it touches the internet is to download the model you chose. The wearable and the watch carry audio to the phone and nothing else. We could not read your conversations if we wanted to; there is nothing on our side to receive them.