TL;DR: Audio never leaves your computer — on any plan, on any system. When translation happens in our cloud, only the transcribed text is sent: never the audio, and we don’t keep the text. We charge for the service, not for your data.
1. Who we are
Lyron is a product by BeCaptain Studio. This policy explains how we handle your data when you use the Lyron desktop app, the lyronapp.com website, and the api.lyronapp.com API.
Contact: hi@lyronapp.com.
2. What we process (and what we don't)
Audio
It never leaves your computer. On no plan, on no system. Capture and speech recognition (Whisper) happen on your machine. We never receive any audio. It's technically impossible: the client doesn't send it.
Transcribed text (Pro only, optional)
If you use Pro mode with Lyron Cloud, the text produced by the transcription is sent to our edge proxy at api.lyronapp.com/v1/translate to be translated. That text:
- Is not stored by us. It's processed in memory and discarded right after we respond. (The translation is done by Google's Gemini API — see section 3.)
- Is not used to train models. Not ours, not anyone else's.
- Is not sold or shared.
The same goes for the AI recap and questions about your history (Pro): the text of the session you picked travels the same way, under the same rules — processed and discarded.
The 7-day free trial is the whole of Pro for a week: while it runs, exactly what is written here applies.
The built-in local translation engine, when you use it, keeps the text on your machine too — then nothing is sent. It exists on Mac and, on Windows, on PCs whose graphics card runs the translator (the app tests this on its own, the first time). On other PCs translation runs in the cloud, and the paragraphs above describe exactly what is sent.
Account and payment
To activate Pro, we need your email (to send you the magic link to sign in) and whatever Stripe collects to process the payment (card, billing address). We never see your card, only Stripe does.
What we store in our backend (Cloudflare D1):
- Your email
- Stripe customer ID and subscription status (active, canceled, or past_due)
- The license key generated for your app (JWT)
- On Pro, the monthly Lyron Cloud usage count: how many translations and recaps, the volume processed and the language pairs — never the text
- The date your account was created
7-day free trial (optional)
You can try Pro for 7 days with just your email, no card. You ask for it inside the app; we send a link to confirm the address, and the trial starts when you click it. When it ends, the app goes back to the free plan on its own and nothing is charged. If you don't ask for the trial you don't need an account: the free plan stays as it always was.
What we store about someone who asks for the trial:
- The email
- The app language and the computer's time zone (the trial ends in the small hours, so it doesn't cut you off mid-use)
- Which screen of the app the request came from, and the dates of the request and of the start of the trial
- Whether you ticked the box to receive news — it comes unticked
- During the trial, the same Lyron Cloud usage count that applies to Pro — never the text
The trial emails are four, and that's it: the link to start, one the next day with what you can do with Pro, a notice before it ends and one when it ends. Lyron news only goes to people who ticked the box. A request nobody confirms is deleted after 30 days.
Download link by email and a conversation with the founder (optional)
If you open the site on a phone (or on a computer Lyron does not run on yet), you can ask for the download link by email, to open on your computer. We keep the email, the page language, the kind of device (phone, Intel Mac or another system) and whether you ticked the news box — it starts unticked. We send one email, with the link. Without the box ticked, the address is deleted after 30 days.
After you use Lyron on 3 different days, the app may ask — only once — whether you would join a 15-minute conversation with the founder, in exchange for 3 months of Pro. If you say yes, we keep the email and the app language only to set up the conversation: we send a confirmation, the founder writes to you, and the address is deleted after 90 days. No account is created because of it.
App telemetry
No third-party analytics tools. No Mixpanel, no Amplitude, no PostHog. The app only uses the internet in these situations:
- downloading the models (the audio one and the local translator), from Hugging Face — at install time or when you ask;
- asking whether a new version exists — on launch and every 6 hours, on any plan;
- on Pro (and in the free trial): validating the licence (on launch and every 6 hours, with the app version and the system), translating text on Lyron Cloud and the AI recap of your history;
- when you ask: signing in (the email goes to the link sender), asking for the free trial (the email, the language, the time zone and the news box), agreeing to the conversation with the founder (the email and the language), opening the subscription portal and seeing the prices in the Pro invitation;
- and the usage measurement below — if it is on.
And it measures its own use, as leanly as we could make it. Without what was translated, without the language and without any identifier of yours, the app reports:
- when a translation session works — the first caption appears —, with the translator and the system;
- when a session doesn't work, with the reason as a fixed word (for example: the audio model wasn't downloaded, or the audio couldn't be opened);
- when the session ends, how long the captions took — the median, the worst case and which time range they fell in —, with the system, the audio model and the translator; and, if Lyron Cloud stopped partway through it, why (as a fixed word) and whether the local translator took over;
- when the Pro invitation appears, when you click to subscribe and when you ask for the free trial — from which screen and whether the link went out, never the email;
- the end-of-session card — at most once a day, when you stop a translation of 3 minutes or more —: which card appeared (the recap, for Pro, or the invitation), the plan and what you chose; and the invitation to talk with the founder: whether it appeared and whether you said yes — never the email;
- if you answer the "how was the translation?" card — shown at most 3 times, when you stop a translation —, the rating, the reason and the fix you chose, with the translator and the audio model;
- on Mac, how long the audio model took to transcribe a test sentence that ships inside the app, the computer's memory (in GB) and whether it kept that model;
- on Windows, the result of the graphics card test — whether it runs transcription and the translator, and whether it failed during use —, with the maker (NVIDIA, AMD or Intel, never the model), the memory in GB and the timings;
- each Vision snip: whether it translated, found no text, or at which step it failed — never the text;
- when a model finishes downloading (or fails), with the model name.
A session produces only a few lines — never one per sentence — and the captions never wait on any of them. In Settings → Account → Usage measurement you switch it off, and then none of this leaves your machine.
Why it exists: without it, someone who opens the app and cannot get it configured is indistinguishable from someone translating every day — opposite problems that call for opposite fixes. And caption speed is the product: without measuring it, there is no way to know whether it holds up on your computer.
The update check happens on any plan. We count those requests — it is how we know how often Lyron gets opened. What gets recorded is the count and the country Cloudflare resolves at the edge: no identifier, no IP, nothing that links one request to the next or to you.
The same switch turns that count off. With Usage measurement off, the update request itself says it does not want to be counted — the app keeps receiving updates normally, it just stays out of the opens tally.
Website telemetry
lyronapp.com uses Cloudflare Web Analytics (privacy-friendly: no tracking cookies, no fingerprinting, aggregated and anonymous) and records a few anonymous navigation events: each page opened (the page address and whether you arrived from outside the site — not where you came from), clicks on the download button — with the kind of device (Mac with Apple silicon, Intel Mac, phone, Windows or another system; never the model) —, whether you asked for the download link by email and whether it went out (never the address), opening the subscription page, the start of a checkout and failures to open the payment page (with the technical reason, never the email). Each one carries only the country, which Cloudflare infers at the edge — no cookies, no IP, and no personal data. They exist only to understand the funnel (visits, downloads, subscriptions) and improve the product. They can't identify you individually.
Advertising
The site loads the ChatGPT Ads measurement pixel to measure which ads bring new users. It stores a first-party cookie, __oppref, holding an identifier for the ad click, and records page views, downloads, checkout starts and completed subscriptions. When a subscription completes, we send OpenAI your email as a SHA-256 hash, never the address itself, so the purchase can be attributed to the ad.
The app still has no ad pixel and no third-party analytics tools. To opt out in this browser:
3. Where the data lives
Infrastructure:
- Cloudflare (Workers, D1, KV): global edge, data at rest in the region closest to the user.
- Stripe: payments (on their servers, under their own privacy policy).
- Resend: sending transactional emails (magic link, receipts, the free-trial emails, the download link and the confirmation of the conversation with the founder).
- Google (Gemini API): processes the text of Lyron Cloud and of the AI recap (Pro), called from our server. We use the paid tier of the API: on it Google does not use the text to train or improve its products, and keeps logs for a limited period only to detect abuse, as the API terms state. On our side, the text is not stored.
- Hugging Face: where the app downloads the models from. Like any website, it receives the download request.
- OpenAI: on the website only, the ad measurement described under Advertising.
We don't run our own data center. Nothing goes to third parties beyond these.
4. Your rights
At any time, you can:
- See your data: ask at hi@lyronapp.com and I'll reply within 7 days.
- Delete your account: canceling in the Stripe Customer Portal removes the subscription, and an email request wipes your email and license key. Stripe keeps tax history because the law requires it. If you only did the free trial: an email request deletes the email and the trial record. If you only asked for the download link or the conversation: an email request deletes the address right away.
- Export your data: anytime, by email.
- Fix your email: straight to me by email.
LGPD (Brazil) and GDPR (EU) apply. We're a small company, but we treat your data the way we'd want ours treated.
5. Cookies
The site uses a single technical cookie for your language preference (PT-BR, EN, or ES). No tracking, no third parties.
The /upgrade page uses sessionStorage to carry your email between the form and the Stripe checkout. It's cleared when you close the tab.
6. Children
Lyron isn't aimed at anyone under 13. If you're a guardian and you've found that your child has an account, write to hi@lyronapp.com and we'll delete it.
7. Changes to this policy
If we change anything material, I'll email everyone with an active account before the change takes effect. The version history stays visible on the changelog.
8. Contact
A question, a request, or a complaint: hi@lyronapp.com. A human (not a bot) replies within 48 business hours.