Skip to content

Data Privacy Policy

Effective Date: October 5, 2026
Last updated: October 4, 2026
AI consent version: ai-consent-v1

This policy is a supplement to our Privacy Policy. It is published as a separate document because ReCut handles recordings of people, and it gives the full account of that: what leaves your iPhone for an AI edit, the AI consent you see in the app word for word, every company that processes data for ReCut, and how long each one keeps it. It also describes these practices in the terms that laws about voice recordings, biometric information and AI disclosures use. Where this policy and the Privacy Policy overlap, they say the same thing; where they appear to differ, the more protective statement applies.

1. Scope and Who We Are

This Data Privacy Policy explains how Novaire Capital Corporation ("we", "us", "our"), a company based in Ontario, Canada, collects, uses, shares and protects the content of your recordings through ReCut - AI Video Editor ("ReCut"), our video editing app for iPhone. It applies to everyone who uses ReCut, wherever they live.

ReCut records or imports talking videos, finds the dead air, filler words, retakes and tangents in them, and lets you review every cut and export the finished video. Only the audio track of your video leaves your iPhone. Video frames, photos, thumbnails and your project library never do. This policy explains exactly what that means.

For questions about this policy, contact us at support@userecut.com, which reaches our Privacy Officer.

2. What "Content Data" Means Here

"Content data" is what your recordings contain and what ReCut makes from them:

  • the picture: video frames, photos and thumbnails, which show your face and the faces of anyone else on camera, and wherever you are;
  • the sound: the audio track, which carries your voice and the voice of anyone else who speaks;
  • the words: the transcript of what is said, with word timings and filler-word flags, and the captions and text you add;
  • the edit: the cut plan, the list of suggested cuts with a reason for each, and your project; and
  • your feedback on a cut, and a transcript only if you choose to include it.

The picture never leaves your iPhone. The sound and the words leave it only for an AI edit, after you consent, as the sections below explain.

3a. What ReCut shows you before any audio is sent

ReCut asks for your agreement on its sign-in screen, before any audio leaves your iPhone. The line under the sign-in buttons reads "By continuing you agree to the ReCut Agreement.", and continuing agrees to it. One tap on that line opens the ReCut Agreement, which shows this text, word for word, for consent version ai-consent-v1:

How ReCut's AI works

ReCut uses AI to find the dead air, filler words, retakes and tangents in your videos.

  • Only the audio track of your video leaves your iPhone. Video frames, photos, thumbnails and your project library never do.
  • Deepgram turns the audio into text. OpenRouter passes the transcript to Anthropic's AI models, which read it to find retakes and tangents, only when those cuts are on.
  • The audio is deleted right after transcription and is never kept longer than 48 hours. Transcripts and cut plans are deleted when your iPhone receives them, or within 7 days if it never does. Backups are kept 7 days. Anthropic may keep the transcript text it receives for up to 30 days, and does not use it to train its models.
  • Deepgram is told not to use your audio to improve its models. ReCut does not train AI models on your videos, audio or transcripts.
  • AI cuts can be wrong. Review your video before you share it; every AI cut can be restored.

You can turn AI processing off in Settings at any time. Turning AI processing off stops new AI edits and cancels AI edits that haven't finished; their audio and transcripts are then deleted. It doesn't recall data already processed, which is deleted on the schedule above. Manual editing keeps working.

You can turn AI processing off at any time in Profile → Settings → AI processing, and back on there. While it is off, nothing is sent and ReCut still works: you can record, import, edit by hand, use Quick cut (the on-device silence cut) and export. If your agreement couldn't be recorded when you signed in, or the text has changed since you agreed, ReCut shows it on a screen of its own before any audio is sent, where you can agree or choose Not now. Our servers check your consent again before every AI edit, so no edit can start without it.

3b. What the AI does

When you ask ReCut to cut a project, it looks for four kinds of material you probably want removed:

  • Silences and dead air. Your iPhone finds these itself, from the loudness of the audio and, in an AI edit, the transcript's word timings. Quick cut finds them from loudness alone and sends nothing.
  • Filler words such as "um" and "uh". Deepgram marks them while it turns the audio into text.
  • Retakes, false starts and repeats. Anthropic's AI models read the transcript and point out the takes you repeated, so ReCut keeps one complete take of each line, usually the last.
  • Tangents and asides. Anthropic's AI models also point out passages that wander off the topic.

The transcript text reaches Anthropic's AI models through OpenRouter, a service that passes it to Anthropic and returns the answer.

Each kind can be turned on or off in Profile → Settings → AI cuts, where retakes appear as "Bad takes". Deepgram receives the audio for every AI edit, whichever kinds are on. OpenRouter and Anthropic are used only when retakes or tangents are on when you start the edit; if both are off, neither receives anything. Every cut has a reason you can see in the editor, and you can restore any of them. AI cuts can be wrong. Review your video before you share it; every AI cut can be restored.

ReCut asks Deepgram to transcribe English, so ReCut's AI works best with English speech. Other languages may be transcribed poorly, and when a transcript is too unclear to rely on, ReCut skips the retake and tangent cuts.

3c. Providers, not models

We name our AI providers, not the models they run. Which model we use is a setting on our servers. It can change without changing what is sent or who receives it. A change of provider, or of what is sent to one, changes this policy, the Privacy Policy, the Terms of Service and the consent text together, and ReCut asks for your consent again before any audio goes to the new provider.

4. Content Data We Collect, and What Never Leaves Your iPhone

Only the audio track of your video leaves your iPhone. Video frames, photos, thumbnails and your project library never do.

To make an AI edit, ReCut extracts the audio track from the clips in your project, combines it into one compressed, mono audio file with no location data, and uploads it over an encrypted connection to ReCut's storage. The upload holds the sound of your recording and nothing else: no picture, no project title and no file names.

Step What is sent To whom Why
Upload The audio track ReCut's storage (Supabase, Canada) So it can be transcribed
Transcription The audio track, fetched through a short-lived private link that contains your random user ID and the edit's ID, never your name or email Deepgram To turn speech into text with word timings and filler flags
Analysis The transcript text, with no user ID, name, title or date OpenRouter, which passes it to Anthropic's AI models To find retakes, false starts, repeats and tangents, only when those cuts are on
Result The transcript and the list of suggested cuts Your iPhone So the editor can apply them

Your iPhone applies the cuts itself, from your original clips, and the finished video is rendered and saved on the phone. Your iPhone keeps a copy of the transcript inside the project, because captions and the editor use it.

What stays on your iPhone and never reaches us: your clips and camera recordings, every video frame, thumbnails, your projects and their edits, captions and on-screen text, exported videos, and the first name and handle you may give during setup. Projects live only on this iPhone. Deleting ReCut deletes them. ReCut keeps no copy of your projects or videos on its servers, so it can't restore them.

5. Where Content Data Comes From

Content data comes from you: what you record with ReCut's camera, the clips you choose to import from your Photos library, and what you type into the editor. ReCut reads only the clips you pick. The transcript and the cut plan are made from your audio by the providers in section 8. We don't buy content data, we don't obtain it from anyone else, and we don't combine it with information from other sources.

6. Why We Collect and Use It

We use content data only:

  • to make the AI edit you asked for and deliver the transcript and the suggested cuts to your iPhone;
  • to keep ReCut working and secure: checking each upload against the project limits, preventing abuse, and finding and fixing failed edits, using records that hold no content; and
  • to improve ReCut from your feedback, including a transcript only if you choose to include it.

We don't use content data for advertising or marketing, to build profiles of you or anyone else, or to make decisions about you. ReCut does not train AI models on your videos, audio or transcripts.

7. How It Is Stored and Transmitted

  • On your iPhone: your projects, clips, transcripts and edits live in ReCut's storage on your iPhone, protected by iOS. Exported videos go to your Photos library, where iOS and your own iCloud settings govern them.
  • In transit: every connection between the app, our servers and our providers is encrypted with TLS.
  • On our servers: the audio waits in a private storage bucket, hosted by Supabase in Canada, that only our server functions can read, until it is transcribed. Transcripts and cut plans wait in our database, which the app can't read directly, until your iPhone receives them.
  • To the AI providers: our servers, never your iPhone, send the audio to Deepgram and the transcript through OpenRouter to Anthropic. The keys for these providers exist only on our servers.

No method of storing or sending information is perfectly secure, and we can't guarantee absolute security. The Privacy Policy (section 12) describes the rest of our safeguards and what we do after a security incident.

8. Who Receives Your Data

This section lists every company that processes personal information for ReCut, including the ones that never see content data. For each one it says what the company does for ReCut, what it receives and where it processes it.

In short: Supabase (Canada), Deepgram, OpenRouter, Anthropic, RevenueCat, Superwall, PostHog, Sentry, Meta, TikTok, Appstack, Apple and Google.

Processor What it does for ReCut Data it receives Where
Supabase Hosts ReCut's database, sign-in, file storage for the temporary audio upload, and server functions Account (email, sign-in provider identity; with Google, the name and profile picture link Google shares), profile (consent answers, notification choices; an age range only for accounts that gave one before October 2026), the audio track until it is transcribed, transcripts and cut plans until your iPhone receives them, usage and subscription records, push tokens, feedback Canada
Deepgram Turns the audio track into text with word timings and filler-word flags The audio track of the project being edited, fetched through a short-lived private link that contains your random user ID and the edit's ID United States
OpenRouter Passes the transcript text to Anthropic's AI models and returns their answer, only when retake or tangent cuts are on The transcript text of the project being edited United States
Anthropic Its AI models read the transcript, received through OpenRouter, to find retakes, false starts, repeats and tangents, only when those cuts are on The transcript text of the project being edited, through OpenRouter United States
RevenueCat Processes App Store subscriptions, restores purchases and reports subscription events User ID, purchase history; for people who allow tracking, device identifiers (the advertising identifier, the identifier for vendors, IP address and device model) and ad-campaign details (the Meta SDK's install identifier, Appstack's campaign, ad and click identifiers), so subscription events can be matched to the ads that led to them United States
Superwall Shows the subscription screen and runs paywall experiments User ID, plan attributes (plan, trial status, where the subscription screen was opened, app build), paywall views and purchase outcomes, and the device details its SDK reads (identifier for vendors, device model, locale, time zone, App Store country) United States
PostHog Product analytics, so we can see which features work User ID and in-app events without personal content (including your setup answers), device model, app version, language and time-zone settings; IP addresses are discarded United States
Sentry Crash and error reports, so we can fix problems User ID, device model, iOS and app version, error details with personal details removed; never screenshots or the screen's contents. Error reports from our servers carry no user ID United States
Meta Measures which Meta ads brought people to ReCut (people who allow tracking, only) Only for people who allow tracking: user ID, a registration event, device identifiers including the advertising identifier; subscription events arrive from RevenueCat Global (Meta's own infrastructure)
TikTok Measures which TikTok ads brought people to ReCut (people who allow tracking, only) Only for people who allow tracking: user ID, install and app-open events, registration, trial and subscription events, App Store purchase events, device identifiers including the advertising identifier Global (TikTok's own infrastructure)
Appstack Measures which ad campaigns brought people to ReCut (people who allow tracking, only) Only for people who allow tracking: user ID, install, first-open and sign-up events, device and install attribution data; subscription events arrive from RevenueCat. Apple's aggregate ad-attribution reports, which identify no one, reach it for every install United States (on Google Cloud)
Apple Sign in with Apple, App Store purchases, push notifications and ad-attribution postbacks Your Apple sign-in identity, purchases, a push token Global (Apple's own infrastructure)
Google Sign in with Google Your Google sign-in identity (email, name and profile picture link) Global (Google's own infrastructure)

8a. Infrastructure

Supabase hosts ReCut's database, sign-in, server functions and the temporary storage that holds the audio track until it is transcribed, in Canada. The app never reads the database directly: every request goes through ReCut's server functions, which check who you are and return only your own data. Your projects are not stored here.

8b. AI

  • Deepgram receives the audio track of the project you are editing and turns it into a transcript with word timings and filler-word flags. Deepgram is told not to use your audio to improve its models.
  • OpenRouter and Anthropic receive the transcript text, never the audio, and only when retakes or tangents are switched on for the edit. The transcript text reaches Anthropic's AI models through OpenRouter, a service that passes it to Anthropic and returns the answer. Anthropic's AI models find retakes, false starts, repeats and tangents. Anthropic may keep the transcript text it receives for up to 30 days, and does not use it to train its models. We send OpenRouter an instruction to route the transcript to Anthropic only, never to another AI company.

The audio is deleted right after transcription and is never kept longer than 48 hours. Transcripts and cut plans are deleted when your iPhone receives them, or within 7 days if it never does.

8c. Payments and the subscription screen

  • RevenueCat processes App Store subscriptions, restores purchases and reports subscription events. Apple processes the payment itself; neither ReCut nor RevenueCat receives your card details. When you delete your account, we delete your customer record at RevenueCat.
  • Superwall shows the subscription screen and tests versions of it. It receives your user ID, a few plan attributes, what happens on the subscription screen, and the device details its SDK reads. When ReCut first opens, its SDK also makes an install-attribution check with Superwall (Privacy Policy, section 8e). It is not an advertising network, and we don't turn on its integrations that forward data to advertising networks.

ReCut's Share analytics switch turns off product analytics only. RevenueCat and Superwall still receive the purchase and paywall data your subscription needs.

8d. Analytics and diagnostics

  • PostHog receives in-app events linked to your user ID, never your content, and is set to discard IP addresses. Session replay is turned off. Profile → Settings → Share analytics turns PostHog off.
  • Sentry receives crash and error reports. It never attaches screenshots or the screen's contents, and it is set not to collect personal information by default.

8e. Advertising measurement (people who allow tracking only)

Meta, TikTok and Appstack measure which ads brought people to ReCut. Their SDKs start only for people who allow tracking in Apple's tracking prompt; ReCut doesn't ask anyone's age, so the prompt alone decides. If the person asks ReCut not to track, they never start. None of them ever receives content data: no video, audio, transcript, caption or project title, and no email or name. The Privacy Policy (section 8) explains what they receive and how to opt out.

8f. Platform

  • Apple provides Sign in with Apple, App Store purchases, push notifications and ad-attribution reports (SKAdNetwork). When you delete your account, we revoke your Sign in with Apple authorization.
  • Google provides Sign in with Google.

Apple and Google also handle your Apple Account and Google Account under their own privacy policies.

8g. Where data is processed

ReCut's database is in Canada. Deepgram, OpenRouter, Anthropic, RevenueCat and Superwall are in the United States, and the table shows the regions of the others. Meta, TikTok, Apple and Google use their own infrastructure around the world. Section 13 of the Privacy Policy explains how we protect information that leaves your province or country.

8h. Other disclosures

We may also disclose personal information when the law requires it, to protect people and ReCut, in a business transfer, or with your consent, as section 7c of the Privacy Policy describes. A successor is bound by this policy, and any new use of content data needs the notice or consent the law requires.

8i. How this list changes

Every SDK in ReCut and every provider that receives personal information is on this list. When we add one, we add it here, with what it does, what it receives and where, and we change the "Last updated" date at the top. We publish the updated policy before or with the release of ReCut that uses the new company, never after.

9. What We Do Not Do

  • We don't sell content data, or any other personal information, for money or anything else of value.
  • We don't use your videos, voice or transcripts for advertising, and no advertising network receives them.
  • We don't train AI models on your content. ReCut does not train AI models on your videos, audio or transcripts. Deepgram is told not to use your audio to improve its models.
  • We don't create biometric identifiers. ReCut does not create voiceprints or any other biometric identifier from your voice. ReCut doesn't recognize faces or voices, and it doesn't use your recordings to identify or verify anyone.
  • We don't publish anything. ReCut never posts, shares or shows your videos to anyone; you decide where your exports go.
  • We don't keep your recordings. ReCut keeps no copy of your projects or videos on its servers, so it can't restore them.
  • We don't listen to your audio or read your transcripts as part of running ReCut. A person reads a transcript only when you include it with feedback, or when the law requires it.

10. Retention and Deletion

What Where How long
The audio track ReCut's storage (Supabase, Canada), then Deepgram while it transcribes Deleted right after transcription; never kept longer than 48 hours
Transcript and cut plan ReCut's database Until your iPhone receives them; at most 7 days
Transcript text sent through OpenRouter to Anthropic Anthropic Up to 30 days
Transcript text passing through OpenRouter OpenRouter Under OpenRouter's own retention policy
Usage records (edits, durations, clip counts, costs; no content) ReCut's database 13 months
Subscription events ReCut's database 13 months
Cut feedback ReCut's database 12 months; a transcript you choose to include, 30 days
Push tokens ReCut's database Until they stop working, or 180 days without being refreshed
Rate-limit counters (your user ID, or a salted hash of your IP address, never the address itself) ReCut's database 48 hours
Account, profile, consent records, subscription status ReCut's database and sign-in service Until you delete your account
Deletion receipt (no personal data) ReCut's database 24 months
Operational alerts and webhook records ReCut's database 90 days
Database backups Supabase 7 days
Projects and videos Only your iPhone Until you delete them, your account (after you confirm) or the app
Copies held by PostHog, Superwall, Sentry, Meta, TikTok and Appstack The processor Under that processor's own retention policy

The AI data in short:

  • The audio is deleted right after transcription and is never kept longer than 48 hours.
  • Transcripts and cut plans are deleted when your iPhone receives them, or within 7 days if it never does.
  • Anthropic may keep the transcript text it receives for up to 30 days, and does not use it to train its models.
  • Deleted data can remain in database backups for up to 7 days before it is gone.

Deleting your account deletes whatever is still on our servers straight away, and turning AI processing off deletes the audio and transcripts of edits that haven't finished. Our Delete Your Data page describes every way to delete your data, step by step.

11. Your Rights Over This Data

Wherever you live, you can:

  • turn AI processing off at any time in Profile → Settings → AI processing. Turning AI processing off stops new AI edits and cancels AI edits that haven't finished; their audio and transcripts are then deleted. It doesn't recall data already processed, which is deleted on the schedule above. Manual editing keeps working.
  • confirm whether we hold content data about you and get a copy of it. Because transcripts and cut plans are deleted from our servers once your iPhone receives them, we usually hold none; the copies are in your projects.
  • receive the list of the companies that received your data. Section 8 is that list.
  • delete your data, in whole or in part, in the app or by email (section 10).
  • not be treated differently for using any of these rights. Some features need certain data, though: AI edits need your AI consent.

How to use them. Use the in-app controls above, or email support@userecut.com from the email address on your ReCut account, at any time, including long after you stopped using ReCut. We verify a request by matching it to the account, and we may ask one follow-up question if that isn't enough. You may use an authorized agent; we will ask for proof of the agent's authority. We answer within the time the law that applies to you sets (the Privacy Policy, section 14, lists them), and we don't charge for a request unless the law allows it for requests that are clearly excessive.

Appeals. If we decline a request, we explain why in writing. You can appeal by replying to that answer, or by emailing support@userecut.com with the subject "Privacy appeal". We answer appeals in writing within the time your law sets. You can also complain to a privacy regulator at any time, as section 14 of the Privacy Policy describes.

12. Changes to This Policy

The in-app consent text, this policy, the Privacy Policy and the Terms of Service describe the same processing and change together. If we change the AI providers, the kind of data sent or how it is used, we publish a new consent version and ask for your agreement again before any more audio is sent. For other changes, we update this policy and the "Last updated" date, and tell you in the app before a material change takes effect.

13. Contact Us

For questions about this policy, or to use any of the rights in section 11, email support@userecut.com, which reaches our Privacy Officer.

Novaire Capital Corporation (Ontario, Canada)
Email: support@userecut.com (support, privacy requests and the Privacy Officer, copyright notices)