On-device AI, built to stay out of the way
nooklet uses AI sparingly, only where it earns its place, and all of it runs on your phone. No cloud service ever sees your child’s audio or words.
Transcription, on your device
Recordings can be transcribed automatically so your journal stays searchable, using small speech models that run on the phone. Audio never leaves the device to be transcribed, and the transcript is editable: a tool, not a source of truth.
Gentle, labelled suggestions
Where nooklet offers a suggestion, it is clearly marked as AI, comes with a plain-language “why this?”, and is supportive rather than directive. It surfaces what it sees; it doesn’t tell you what to do.
Patterns that stay yours
Pattern-finding is rule-based correlation with a little on-device learning, not a global model. It runs on your family’s data alone, describes rather than predicts, and never compares your child to anyone else. There is no cloud model analysing your child’s data.
What runs where
| Feature | Where it runs | What leaves the device |
|---|---|---|
| Transcription | On device | Nothing |
| Search | On device | Nothing |
| Suggestions | On device | Nothing |
| Pattern-finding | On device | Nothing |
| Family sync | Encrypted before leaving | Encrypted blobs only |
What we will never do
We do not train models on your child’s data, do not aggregate across families, do not send the journal to cloud AI, and do not generate child-facing content. Every AI surface is labelled, per EU AI Act transparency. See the full privacy notes.
On the way
Plain-language journal search (ask a question in your own words and let nooklet find the moments) is in progress, and runs on-device like everything else.