On-device AI runs on your own computer, so work data is processed locally instead of being sent to an AI provider. It is the strongest privacy option for workplace AI, at the cost of some capability compared with the largest cloud models. Good tools let you choose: local by default, cloud when you decide it is worth it, and a clear record of what was sent.
Why it matters at work
- Meeting content, customer details and internal decisions are sensitive.
- Many organizations restrict sending data to outside AI services.
- Employees trust tools more when their raw activity stays with them.
What on-device AI can do well today
- Summarize notes and meetings
- Extract tasks, decisions and facts
- Search your own documents
- Text recognition on scans
Where it still struggles
- Very long, complex reasoning
- Some languages
- Older or low-memory hardware
How to judge "private AI" claims
| Question | Good answer |
|---|---|
| What leaves the device? | A specific list, ideally visible to the user |
| Does it work offline? | Yes, for core features |
| Can I choose the AI engine? | Yes, local or cloud, per organization |
| Is my data used to train models? | No, unless you opt in |
Reattend's approach
Reattend writes memories with Rabbit, Reattend's own model, on the computer by default. Rabbit is still in training. A person using Reattend on their own can use their own AI provider key instead, and managed AI through reattend.com is a pilot Reattend enables on request; with either, text from activity is sent to that AI service for processing. Raw capture is stored only on the device, only derived knowledge is shared with the organization, and a ledger shows what left the Mac.
Frequently asked questions
What is on-device AI?
AI that runs on your own computer or phone instead of a remote server. Your data is processed locally and does not need to be sent to an AI provider.
Is on-device AI as good as cloud AI?
Not yet for every task. Smaller local models are improving quickly and handle many tasks well, but the largest cloud models are still more capable on complex reasoning.
How can I tell if a tool really processes data on-device?
Read its privacy documentation for exactly what leaves the device, check whether it works offline, and look for a log of what is sent.