The transparency page
How it reads.
And what it never does.
This page exists because "trust us" isn't a feature. Here is exactly what happens to your clip, which services see it, what the welfare gate checks, and how we test all of it before every release.
The journey of one clip
1. You film up to fifteen seconds. The clip stays on your phone until you ask for a read.
2. The clip is uploaded to our backend (storage and database run on Supabase; the worker that coordinates the pipeline runs on Fly.io).
3. Google's Gemini model reads the clip. It returns a structured description of the body language it sees. Under Google's commercial API terms, your content is not used to train their models.
4. The welfare gate runs on that reading — before anything playful is allowed to happen. The gate's rules are below.
5a. If the read is calm, clear, and playful: a short voice line is written and sent to ElevenLabs to be synthesized. ElevenLabs receives only short text strings — never your video, never audio from your clip.
5b. In every other case: no voice line. If the reading included signs we want a professional to look at, you see a calm card with a concrete next step — call your vet, or find a veterinary behaviorist. It describes what was seen; it never names a condition.
6. The result is stored under your account — your clips, readings, and voiced videos — where you can inspect or delete them anytime from inside the app.
That is the complete list of services that touch your content. Each one is contractually bound not to train on it and not to use it for advertising. The same list, with links to each provider's privacy terms, is in section 4 of our Privacy Policy.
What the welfare gate checks
The gate asks three questions of every reading, in order:
Is there any sign a professional should see? Signs associated with pain, distress, or an emergency stop everything. No voice line, no playfulness — you get the calm escalation card instead. PawQuirks is never playful about a pet that looks uncomfortable.
Is the read clear enough to act on? If the clip is too short, too dark, too shaky, or the pet isn't really in frame, PawQuirks abstains. It tells you why and suggests how to get a better clip. It does not fill silence with invention.
Is the mood actually playful? Only then does a voice line get written — and it's written from what the reading actually saw, not from a joke template.
These rules are not a settings toggle. There is no way to turn the gate off — not for you, and not for us.
How it's tested
Before any update ships, the welfare logic is validated against a 60-scenario welfare suite: a fixed set of situations covering emergency signs, subtle discomfort, unclear clips, and happy play. The suite checks three things — that emergency signs always escalate, that an uncomfortable pet is never voiced, and that unclear clips end in an honest abstain instead of a guess.
The same rules are implemented in three places (our research prototype, our server, and the app itself) and all three are tested against the same suite, so they can't drift apart. A release that fails any scenario does not ship. The suite was most recently run in June 2026, ahead of launch.
What it never does
It does not give health verdicts. PawQuirks reads body language. It is not a veterinary tool, and it cannot tell you whether your pet is healthy. It can be wrong. If something about your pet worries you, call your vet — don't wait on an app.
It does not pretend the voice is real. The voice line is AI-generated, written from the reading. The pet is one hundred percent real; the words are not.
It does not sell or advertise with your data. No ads, no data sales, no third-party analytics SDKs. The app is free, with no subscription and no in-app purchases in this version.
It does not keep what you delete. Deleting your account from inside the app (Home → gear icon → Settings → Account) removes your pets, clips, readings, voiced clips, and activity events from our servers. One exception, kept on purpose and described in the Privacy Policy: an anonymous, content-hashed cache of synthesized audio that is not tied to you and contains no personal information.
Questions this page didn't answer? Email hello@pawquirks.app — a real human reads it, and welfare concerns get a same-day reply.