Your texts do not travel.
Last updated: November 4, 2026. Controller: José Luis Saorín Ferrer, Tenerife (Spain).
What this is
No account, no name, no texts
Touches the network (1): the model downloads
To speak, Yappy needs its voice model, and it downloads it once: about 200 MB on the phone, about 398 MB on the computer. It comes from modelos.yappy.joseluissaorin.com, a Cloudflare R2 store. On iPhone and iPad it usually arrives with the app itself: the system fetches it from that store with Background Assets, and from nowhere else. If it didn't arrive that way, the app downloads it on launch — on its own if you're on wifi, otherwise when you ask it to. On Android the download starts by itself the first time you open the app; on the computer, when you ask for it.
If our store is down, the app tries the public copy of the model on Hugging Face (huggingface.co, which serves the files from its own distribution network). Only then, and only for the files that didn't arrive from our store.
On the phone, the first run includes the parrot's voice recorded in Spanish and English; in the other 29 languages, those recordings are downloaded from the same store when the app starts in your language, or when you pick one.
Turning voice notes and other audio into text needs a second model, the ear — the transcription model (Parakeet, about 670 MB). That one downloads only when you ask, and it comes from Hugging Face (huggingface.co). The OCR, on the other hand, ships inside the app and downloads nothing.
In all these downloads, the server sees your IP address and which files you ask for, as with any download; no identifier of yours travels with them, nor anything you read. We keep no log that identifies you on our store. After that, the voice and the transcription work in airplane mode forever.
Touches the network (2): the links you share
If you share a link or a video, your device visits that address to fetch the text or the transcript, just as your browser would. The visit comes from your device, not from a Yappy server, and the site receives what any browser would send it: your IP address and your device type. No history is kept anywhere outside your device.
Third-party services. A few links are fetched through public third-party services. Hacker News threads are requested from the Algolia API (hn.algolia.com), which receives the thread number. Long X articles (not regular posts, which are fetched from X itself) are requested from api.fxtwitter.com, a public service independent of X, which receives the post number. And on mobile, if a link fails and you tap the “From the archive” button, Yappy asks the Internet Archive's Wayback Machine (web.archive.org) for the saved copy, and it receives the full address of the link. As with any request, these services also see your IP.
The invisible browser. On iPhone, iPad and Android, if the direct download doesn't get the article (the site refuses it, a cookie notice is in the way, or the text only appears once the page runs its JavaScript), Yappy opens the page in an invisible browser on the device itself (WKWebView on iOS, WebView on Android) and, if a cookie notice pops up, accepts it for you to reach the text. Once it accepts, as in any browser, the site can set its cookies, including advertising ones. Those cookies stay inside Yappy, separate from your browser's. On iPhone and iPad they live only in memory and vanish when the app is closed. On Android, Yappy clears the site's cookies as soon as the page finishes loading; if the attempt fails, or the page jumps to another domain, some may remain stored in the app until you delete its data or uninstall it.
Paywalled articles. Yappy does not sneak past paywalls. If you have a newspaper subscription, on mobile you can sign in once from Yappy: the newspaper's own page opens, you enter your username and password there, and Yappy never reads or stores them. The session stays on your device alone, and from then on the invisible browser opens that site's articles with it (cookies accepted there are stored with the session). You can delete sessions whenever you like in Settings, under “Subscriptions logged in”.
Scientific articles. If an article with a DOI is paywalled at the publisher, Yappy asks Unpaywall, a free service from OurResearch, whether a legal open-access copy exists, for instance in a university repository. If it does, Yappy reads that copy and, on mobile, tells you where it came from. Unpaywall receives only the article's DOI and a contact address its API requires — Yappy's, not yours; like any request, it also sees your IP. Your device downloads the copy afterwards, like any other link.