Talvor Privacy Policy

The short version

• There is no Talvor account, no sign-in, and no backend that stores your audio, your transcripts, or your call notes.
• By default, everything a live call needs — transcription, telling speakers apart, matching a voice to your enrollment, and writing suggested questions — runs on your device using Apple's Speech, Foundation Models, and Core ML frameworks. It works offline.
• If you turn on Cloud questions, the text of the recent conversation (not audio) passes through Meyer App Studio's relay to Anthropic's Claude, which writes the suggested questions. The relay stores none of it. See Cloud questions below.
• A saved transcript, a saved recording, your voice-enrollment sample, and your call history stay in Talvor's storage on your device. They leave your device only when you explicitly export or share them yourself.
• Talvor collects no analytics, no crash reports, and no advertising identifiers. There is no tracking of any kind inside the app.

What Talvor processes, and where

The one-time downloads

On first use, Talvor downloads its on-device models so they can run offline afterward:

• Apple's speech and language models, delivered by iOS itself the same way any Apple Intelligence feature installs its models.
• The speaker-separation and voice-matching models, an open-source Core ML model package downloaded once from Hugging Face's model hosting.
• A second speech model for sharper transcription (about 450 MB), downloaded once from Hugging Face's model hosting, only over Wi-Fi.

These downloads are the model files themselves — not your audio, your speech, or anything about your calls. Once downloaded, they run fully offline for every call after that.

Cloud questions (optional, United States)

Cloud questions comes with Talvor 2.0 (in beta testing now; earlier versions don't have it). It is off unless you turn it on, and Talvor asks first, explaining what is sent. It is available only on the United States App Store. If your App Store country changes, it turns off.

• What is sent, while you're listening: the text of about the last seven and a half minutes of the conversation, labelled by speaker (for example "You" and "Dana"), the names of the people on the customer side, the questions already suggested in that conversation (so they aren't repeated), and the subject and facts Talvor has already noted (so they aren't noted twice).
• Also sent with each request: proof from Apple that this is a genuine Talvor install and, if you subscribe, of your Talvor Pro subscription (signed by Apple; no payment details), and your App Store country. These let our relay serve only subscribers and free-trial users in the United States.
• Never sent: audio, your voice-enrollment sample, saved calls, meeting notes, or anything from Call History.
• Who handles it: Meyer App Studio's relay (hosted on Netlify) passes the text to Anthropic's Claude, which writes the suggested questions, the conversation's subject and the facts shown under "What it's picked up", and returns them to the app. Anthropic does not use this text to train its models. Under its standard terms for API customers, Anthropic deletes inputs and outputs within 30 days, except where it must keep them longer to enforce its usage policies or to comply with the law. We have asked Anthropic for zero data retention for Talvor and will update this page if it applies.
• What our relay keeps: none of the conversation. It logs only the time a request took, its size in tokens, and how many questions came back, and it counts requests per install to prevent abuse, under a code that changes every day and is deleted after a day, never the App Store identifier itself.
• Consent: the people you're talking with are part of what is sent. Use Cloud questions only when everyone in the conversation is comfortable with that.
• Turning it off: Settings (the gear on the main screen) › Questions. With it off, Talvor goes back to writing questions on your device, offline included.

The one fallback worth being direct about

Recording and consent

Saving a transcript or an audio recording is off by default. Turning either on, or starting a call, shows a consent notice reminding you that recording-consent laws vary by location and that you're responsible for getting agreement from everyone on the call. Talvor keeps a timestamped record of when you affirmed that notice, stored locally with the call and included if you export it — so the record travels with the words it covers.

What we don't do

• We don't store your calls, transcripts, or notes on any server. With Cloud questions on, conversation text passes through our relay and is not kept there.
• We don't use analytics, crash reporting, or advertising SDKs of any kind.
• We don't sell or rent your data. With Cloud questions off we never have it; with it on, the conversation text goes to Anthropic only to write your questions.
• We don't require an account, an email address, or a phone number to use Talvor.

Deleting your data

Everything Talvor stores lives in the app's local storage on your device. Deleting a saved call from Call History removes its transcript and recording from your device. Deleting the app removes everything, including your voice enrollment. There is nothing on any Meyer App Studio server to separately delete: our relay keeps no conversation text, even with Cloud questions on.

Children's privacy

Talvor is a professional tool for field sales conversations and is not directed at children. We do not knowingly collect information from anyone under 13.

Changes to this policy

If this policy changes in a way that matters — most likely because a new feature changes what runs on-device versus off — we'll update this page and the "Last updated" date above. We won't make a material change without updating it here first.

Contact

Talvor is published by Meyer App Studio, LLC. Questions about this policy or how Talvor handles data can go to info@meyerappstudio.com.