Privacy Policy

Last updated: August 27, 2026

Cadence is a live captioning app for deaf and hard-of-hearing users. It runs entirely on your iPhone.

This policy is short because there is very little to describe. The Cadence app has no user accounts, no servers, and no analytics. We do not collect, receive, store, or transmit your personal information from the app. We could not identify you from your use of it if we tried.

This marketing website is a separate matter, and a much smaller one: it counts pageviews, and only if you agree. That is described under "The website" below, and in full in our Cookie Policy. Nothing on the website has any connection to what happens inside the app.

The short version

• Your audio never leaves your device.
• Your captions never leave your device.
• We have no backend, so there is nowhere for your captions or audio to go.
• The only network connection the app makes is to download speech models, and that request contains nothing about you.
• The website counts pageviews, sets no cookies unless you accept, and knows nothing about your captions.

What Cadence processes on your device

To caption speech, Cadence needs access to certain things on your iPhone. All of it is processed locally and none of it is sent to us.

Microphone audio

Cadence records audio through your microphone while a captioning session is active, and converts it to text using speech models that run on your device. Audio is held in memory only for as long as it takes to caption it, and is then discarded. Audio is never written to a file, never uploaded, and never used to improve our models or anyone else's.

Speech recognition is pinned to on-device processing. If an on-device speech engine is unavailable, Cadence will report an error rather than fall back to a cloud service.

Captions

Captions produced during a session are saved to your device's local storage so you can review past sessions. These files live in the app's private Application Support directory and are accessible only to Cadence. They are included in your iPhone's backups if you have backups enabled, under whatever protection you have configured for those backups. Through iCloud, that is governed by Apple's privacy policy, not ours. You can delete any session from within the app, and deleting the app removes all of them.

Alongside these, Cadence appends a line of numbers to a local file each time a session ends, recording things like how long the session ran and how quickly captions settled. It holds no caption text and no audio, it is written to the same private directory, and it is never transmitted anywhere. Deleting the app removes it.

Captions over other apps

Cadence can keep captioning while you use another app, showing the most recent line in a small floating window. The text in that window is drawn by Cadence on your own screen and is not shared with the app underneath it, which cannot read it. Nothing about this feature sends anything off the device. It is on by default and can be turned off in Settings, which makes switching apps pause captioning instead.

Contacts (optional, off by default)

If you turn on the name-casing feature, Cadence reads the given names of your contacts and uses them to capitalize names correctly in captions, so that a caption reads "Sarah" rather than "sarah." Only given names are read. They are used solely as a spelling dictionary, are never used to identify who is speaking, and never leave your device. This feature is off unless you enable it, and you can turn it off at any time.

Speaker labels

When speaker identification is enabled, Cadence distinguishes between voices in a conversation and labels them anonymously as Speaker A, Speaker B, and so on. No voiceprint or biometric template is stored. The distinction exists only for the duration of a session and is discarded when the session ends. Labels cannot be matched to a person, and cannot be carried across sessions.

Network activity

Cadence connects to the internet for one purpose only: downloading the speech models that make on-device recognition possible. Nothing else the app does uses the network.

Which model you choose determines where that download comes from. The Default and Alternate models are fetched from Hugging Face, where they are published by FluidInference (the maintainers of the open-source FluidAudio project) and by NVIDIA. The Built In models are Apple's, and iOS fetches them from Apple as a system-managed download that Cadence asks for but does not carry out itself. Either way the download happens once and is then cached on your device.

These requests contain no account identifier, no caption text, no audio, and nothing else about you or your usage. They are ordinary file downloads. As with any internet request, the host delivering the file, Hugging Face or Apple, may observe your IP address as a technical necessity of sending it to you, under its own privacy policy; we neither receive nor have access to those logs.

The Cadence app makes no other network requests. There is no telemetry, no crash reporting, no analytics SDK, no advertising identifier, and no cloud fallback for recognition. The analytics described below run on the website only and are not present in the app.

Diagnostic logs

Cadence can export a diagnostic log to help troubleshoot problems. This export is user-initiated (it happens only when you tap the button), and it is handed to the iOS share sheet so that you choose where it goes. Nothing is transmitted automatically.

Diagnostic logs contain pipeline events such as audio format, timing measurements, and error messages. They contain no caption text and no audio. If you choose to send a log to us for support, please review it first; whatever you send, we will use only to diagnose the issue you reported.

What we collect

From the app: nothing.

We have no servers that receive your captions or audio, no database that could store them, and no analytics inside the app that would report them. If you email us for support, we will have whatever you put in that email, and we will use it only to answer you.

The website

This website, cadencecaptions.com, counts pageviews using PostHog, so we can tell which pages are actually read. This is the one place where we collect anything at all, and it is about pages, not people.

Until you accept, analytics run in cookieless mode: nothing is stored on your device and your visit is counted using a privacy-preserving hash that cannot be linked to any other visit. If you decline, it stays that way. If your browser sends a Do Not Track or Global Privacy Control signal, we honor it regardless.

We have turned off the parts of PostHog that would collect more: no autocapture of clicks or form input, no session recording, and no profiles for anonymous visitors. We do not use advertising cookies and we share nothing with ad networks.

The Cookie Policy sets out exactly what is stored and for how long.

None of this touches the app. The website cannot see your captions, your audio, or anything about your use of Cadence, because the app sends it nothing.

Children's privacy

Cadence does not collect personal information from anyone, including children. We do not knowingly collect data from children under 13 because we do not knowingly collect data from anyone.

Your rights

Privacy laws such as the GDPR and CCPA give you the right to access, correct, delete, or port the personal data a company holds about you. We hold no personal data from the app, so there is nothing for us to produce or erase. The data Cadence creates lives on your device and is under your control: you can delete individual sessions in the app, or delete the app to remove everything at once.

The website's pageview analytics are the only data we receive at all. You can withdraw consent at any time by clearing this site's cookies and site data, which resets the banner and removes the analytics cookie. Declining, or never answering, means no identifier is stored in the first place.

We do not sell personal information. We do not share personal information. We have never done either, and we have no advertising or data-broker relationships of any kind.

Changes to this policy

If we change this policy, we will update the date at the top of this page. If a change ever means the Cadence app starts collecting data it does not collect today, we will say so clearly in the app before it takes effect, not quietly in a document you would have to re-read to notice. That commitment is unchanged.

Website analytics were added in July 2026. They are opt-in, they are described above and in the Cookie Policy, and they collect nothing from the app.

The August 2026 revision describes the floating caption window and the local session-metrics file, and names Hugging Face as the host the speech models are downloaded from. Nothing about it widens what Cadence collects, which is still nothing.

Contact

Questions about this policy can go to [email protected].

Cadence is made by Index Labs LLC.