Effective date: June 10, 2026
This policy explains how Tilt Talk uses generative artificial intelligence and the rules for using it. It supplements our Terms of Service and Privacy Policy.
What the AI does
Tilt Talk uses AI to translate what two people say to each other, face to face. Two kinds of models run entirely on your device: a speech-recognition model (NVIDIA Parakeet) that turns your speech into text, and a translation model (Google TranslateGemma) that translates that text into the other speaker's language. On devices that cannot run the translation model, Apple's on-device translation engine is used instead. There is no cloud AI service behind the app.
How your data is used by the AI
The translation model receives only the current utterance — the text just transcribed from what was said — and produces the translation on your device. Your speech and the resulting text are not stored after the translation is shown, are not sent to us or to any AI provider, and are never used to train, fine-tune, or improve any AI model. The original transcript is always displayed next to the translation so both speakers can verify it.
Accuracy and limitations
AI translation is helpful but imperfect. It can mistranslate idioms, slang, regional vernacular, names, and numbers, and output is usually in a standard register rather than a local dialect. Occasionally the model may add a short stray phrase or produce an awkward result. Do not rely on the app for high-stakes communication — medical, legal, financial, safety, or emergency situations — where a mistranslation could cause harm. When accuracy matters, confirm important details or use a qualified human interpreter.
Acceptable use of the AI features
You agree not to use Tilt Talk's AI features to generate, facilitate, or carry out activities that are illegal or that violate the rights and safety of others. This includes, without limitation: unlawful or dangerous activities; content that sexually exploits or endangers children; harassment, threats, defamation, or hate; deception, fraud, impersonation, or the creation of misleading content presented as authentic; and infringement of others' intellectual-property or privacy rights. Because the translation feature is built on Google's TranslateGemma model, your use is also subject to the Gemma Prohibited Use Policy (ai.google.dev/gemma/prohibited_use_policy) and the Gemma Terms of Use (ai.google.dev/gemma/terms).
No human review
Because translation happens entirely on your device, we have no ability to see, hear, monitor, or review your conversations or the AI's output. This protects your privacy, and it also means we cannot moderate AI output in real time. You are responsible for how you use the translations the app produces.
Reporting a problem
If the AI produces a harmful, offensive, or seriously wrong result, please tell us at support@tilttalk.app. You can attach the app's diagnostic log, which contains technical information only and does not include your conversation content.
Changes to this policy
We may update this policy from time to time. When we do, we will revise the effective date above and, for material changes, provide notice in the app.