NuviWith many AI note takers your audio leaves the device, is transcribed on a server, and may be stored or used to improve models. Ask six things of any tool: where audio is processed, which AI provider receives it, how long it is kept, whether it trains models, who can read the notes, and whether you need an account.
A recording of a meeting holds more than the words. It holds voices, names, salaries, health details, client plans and offhand remarks nobody expected to be kept. When an app turns that into notes, the audio usually travels somewhere to be processed. Where it goes and how long it stays are the questions that matter.
This guide gives you six questions to ask of any AI note taker, shows how to find the answers, and then answers each one for Nuvi, including the parts that are less flattering.
What usually happens to audio in an AI note taker?
The common path has four steps:
- Capture. A bot in the call, an app on your computer, or a phone microphone records the audio.
- Upload. The audio is sent to a server. Some products transcribe on the device instead.
- Transcription and summary. A speech model turns audio into text, and a language model writes the summary. These may belong to the vendor or to another AI company the vendor pays.
- Storage. The audio, transcript and summary are kept somewhere: on the vendor’s servers, on your device, or both.
Each step is a point where a product can make a different choice. None of the choices is hidden if the vendor writes clearly. Your task is to read what they wrote.
Six questions to ask of any tool
1. Where is the audio processed?
On the device, or on a server? If a server, in which country or region? On-device processing keeps the audio with you. Server processing lets a product run larger speech models than a phone can.
2. Which AI provider receives it?
Many products do not run their own models. They send audio or transcripts to a transcription or language-model provider. That is a second company handling your meeting. Look for a list of “subprocessors” or “service providers”.
Granola’s help center, for example, says it “passes audio directly from your microphone and system audio to our transcription provider” and that it “does not record or save audio or video at any point during the call”.
3. How long is the audio kept?
Three patterns exist: kept until you delete it, deleted after a fixed period, or deleted once the transcript exists. Kept audio is useful, since you can replay a moment to check the transcript. It is also the thing most worth protecting.
4. Is it used to train models?
Read for the words “train”, “improve our services” and “de-identified”. Otter states that it “uses a proprietary method to de-identify user data before training our models”, and that its AI service providers do not use customer data to train theirs. That is a clear statement of a different choice from a vendor that does not train on customer data at all. Decide which you are comfortable with.
5. Who can read the notes?
Think about three groups: the vendor’s staff, your teammates and admins in a shared workspace, and anyone with a share link. A team product is built for other people to see notes. That is a feature, and also something the people you recorded may not expect.
6. Do you need an account?
An account ties recordings to an email address and often to a calendar. A product that works without one has less to link to you.
How do you read a privacy policy in ten minutes?
You do not need to read every line. Search the page for specific words.
| Search for | What you learn |
|---|---|
| “retain”, “retention”, “delete” | How long audio and transcripts are kept |
| “train”, “improve”, “de-identif” | Whether your content feeds model training |
| “subprocessor”, “service provider”, “third part” | Which other companies handle your data |
| “region”, “stored in”, “transfer” | Where the servers are |
| “sell”, “advertis” | Whether data is used for ads or sold |
| “account deletion”, “erase” | Whether you can remove everything, and how |
Then check two other places:
- The App Store privacy section. Apple requires developers to describe their app’s privacy practices, “including the practices of third-party partners whose code you integrate”, and shows on each product page whether data is linked to you or used to track you. Developers report this themselves, so treat it as a claim to compare against the policy.
- The help center. Product documentation is often more concrete than the legal policy.
Be wary of a policy that only speaks in adjectives. “Bank-level security” tells you nothing. “Audio is deleted after transcription” is a statement you can hold a vendor to.
The other people in the room
Your privacy is half the picture. The people you record did not choose the tool. Recording others may require their consent depending on where you are, so tell people before you record. If you are recording on behalf of a business in the UK, the regulator’s advice is to “tell people why you’re recording, what you’ll use it for, and how long you’ll keep it”. You can only do that if you know the answers to the six questions.
The consent rules by region are in is it legal to record a meeting.
How Nuvi answers each question
Nuvi is the AI note taker for iPhone, iPad and Apple Watch by Ege Beşe. These answers describe how it is built.
| Question | Nuvi’s answer |
|---|---|
| Where is audio processed? | Not on the device. Audio is uploaded over an encrypted connection (TLS) to temporary cloud storage (Google Cloud Storage, EU region) only so it can be transcribed. |
| Which AI provider? | None outside the developer’s own servers. From storage the audio goes to speech and language models that run on servers the developer operates. It is not sent to a third-party AI service. |
| How long is it kept? | The uploaded audio is deleted as soon as the transcript has been returned. A storage rule deletes any leftover upload after one day. Transcripts and summaries are produced on the server and returned; they are not stored on the server. |
| Used for training? | No. Audio and transcripts are not used to train AI models. |
| Who can read the notes? | Notes live on the device and, if you use iCloud, in your own private iCloud, which the developer cannot read. There is no team workspace and there are no shared folders. |
| Account? | No account and no email. The app signs in anonymously in the background with a random identifier. |
The parts worth saying plainly
- Your audio does leave the phone. Nuvi does not transcribe on the device. If your requirement is that audio never leaves the device, Nuvi does not meet it, and an on-device tool is the right choice.
- A cloud storage provider is involved. The temporary upload sits in Google Cloud Storage in the EU until the transcript is returned.
- There are usage statistics. The app collects anonymous usage statistics (TelemetryDeck) and crash counts (Apple MetricKit). Neither is linked to identity. Subscription status is checked through RevenueCat with an anonymous identifier.
- No certifications are claimed. Nuvi makes no compliance claims. What is stated is that audio is sent over an encrypted connection and handled as described above.
- Transcription needs a connection. Recording works offline, and the recording waits on the device until the connection returns.
What stays on the device
Voiceprints, which Nuvi uses to recognize a voice you have named, are computed while the audio is transcribed, returned to the device and never stored on the server. They stay on the device, are not synced to iCloud and are excluded from backups. How speaker separation works is covered in speaker diarization explained.
There are no ads, no ad tracking and no selling of data. “Delete all data” in Settings removes everything: on the device, in iCloud and on the servers.
The longer version is on privacy first, and the legal text is the privacy policy. For side-by-side comparisons, see Nuvi vs Otter and Nuvi vs Granola.
A short checklist before you record anything sensitive
- Find the retention answer. If you cannot, assume the audio is kept.
- Find the training answer, and the opt-out if there is one.
- Know which companies handle the audio.
- Tell the people in the room, and get a yes.
- Delete recordings you no longer need.
