What Vavel reads, stores and shares

Vavel exists to translate your messages. This page lists exactly what that needs.

What Vavel reads

Only the text you send, only in apps you switched on, and only when you press Send.
Vavel does not read apps you did not enable, messages you receive, or text you do not send.
With instant mode on, Vavel also translates your draft after you pause typing, so Send is fast. You can turn this off in the dashboard.

Your username, your email if you gave one, and a hash of your password.
Your settings: apps, languages, tone, mode, glossary and device names.
Usage counts for each day: characters and requests.
Vavel never writes message text to its database or its logs. A short memory cache keeps recent translations for up to 10 minutes and then drops them.

Groq runs the translation model first.
If Groq fails, Anthropic runs it as a fallback.
Vavel sends the text to these providers only to translate it. Links, mentions, emoji codes and code blocks stay on your side and come back unchanged.

Vavel makes a voice model from your sample only after you tick the consent box. Your sample is biometric data.
Vavel keeps no recording. The voice server keeps only the voice model, and Vavel stores the date of your consent.
Groq turns your spoken phrase into text. Vavel translates it and the voice server speaks it in your voice. Every voice output carries an inaudible AI-speech watermark.
Delete my voice removes the voice model and the consent record at once.