On-device or cloud: which should you choose?
Visit Digest can turn your recording into a summary in two ways. In the cloud, where our cloud model does the work, or entirely on your iPhone, where nothing about your visit ever leaves the phone. We tested both on the same nine realistic doctor visits so you can see the trade-off for yourself.
The short answer
The cloud is the most complete. It caught 90% of medicines and 94% of health conditions across all nine visits, and it was the more reliable of the two on long, busy appointments.
On-device is good enough for most visits, and completely private. It caught 83% of medicines and 79% of conditions. It reliably picks up the main reason for the visit, the new or changed medicines and what you need to do next. What it misses tends to be the background: an over-the-counter remedy mentioned in passing, a supplement, or an older piece of history.
What each one captured
Share of the medicines and health conditions said in each visit that ended up in the summary, across all nine visits (two runs each).
Hover or tap a row for the numbers. Tap a method to highlight it.
Visit by visit
Key facts captured in each visit (medicines and health conditions together). Visits run from 5 to 30 minutes, with up to four people in the room. The further apart the two dots, the bigger the gap.
Hover or tap a row for the numbers. Tap a method to highlight it.
Show the numbers as a table
| Visit | Length | On-device | Cloud |
|---|---|---|---|
| Heartburn | 5 min | 67% | 100% |
| Child's ear infection | 7 min | 75% | 100% |
| Low mood | 10 min | 100% | 100% |
| Anxiety | 11 min | 83% | 100% |
| Chest pain | 15 min | 82% | 91% |
| Biopsy results | 19 min | 92% | 100% |
| Falls & medicine review | 20 min | 74% | 87% |
| Pregnancy booking | 20 min | 86% | 77% |
| New patient, many concerns | 30 min | 83% | 100% |
On short and medium visits the two are close, and on one visit (the pregnancy booking) on-device found more medicines than the cloud. The gap widens on long visits with many topics, like a 20-minute medicine review for an older patient.
Which should you choose?
Most complete
- Captures the most detail, especially on long visits with lots of medicines or history.
- Fast: a summary in seconds once your recording is uploaded.
- Your recording is sent securely to our servers to be processed. Read how we handle it in our privacy policy.
Total privacy
- Nothing leaves your iPhone: not the audio, not the transcript, not the summary.
- Gets the main points right: why you went, what changed and what to do next.
- May miss smaller details, like a supplement or an older diagnosis. Check your summary against your notes.
- Needs a recent iPhone that supports on-device AI. It takes a little longer, as your phone does the work.
Want it truly private? Turn on On-device processing and turn off Cloud Sync, both in Settings > Privacy & Security. Your recordings are then transcribed and summarised on your iPhone, and your visits are kept only on that phone. The trade-off is that nothing is backed up: if you lose the phone or delete the app, those visits can't be recovered.
You can switch between them at any time in Settings. Ask AI and visit prep always ask before sending an on-device visit to the cloud. Whichever you pick, Visit Digest summaries are AI-generated and are not medical advice: if something matters, check it with your care team.
How we tested
We wrote nine realistic doctor visits with a clear answer key: the medicines, doses and health conditions actually discussed. They cover a heartburn check, a child's ear infection, low mood, anxiety, chest pain, biopsy results, a falls and medicine review, a pregnancy booking and a 30-minute new-patient visit, with two to four people speaking and sensitive topics included.
Each visit was processed twice by each method, from the same transcript, using the same processing code Visit Digest runs in the app. We counted how many of the answer-key medicines and conditions appeared in each summary, accepting everyday wording (for example "high blood pressure" for hypertension). Both methods completed all 18 runs, and neither added a medicine that wasn't mentioned in the visit.
We wrote real-world touches into these visits, like background sound, interruptions and people correcting themselves, but this is still mostly clean data: scripted, clearly spoken and processed from an accurate transcript. Real appointments are messier, with crosstalk, accents and a phone further from the speaker, so results in your own visits may differ. Timings exclude uploading and transcription; on-device speed depends on your iPhone. We'll update this page as Visit Digest improves.
