Privacy

What fyv hears, and what it keeps.

fyv translates the speech in a call while it is happening. To do that it has to listen to the call. This page says exactly what that means. Last updated 6 October 2026.

The short version

Audio is translated, not recorded. It streams through our relay to the speech vendors and is discarded as soon as the translated speech has been sent back. fyv keeps no audio and no transcripts.
Only while you press Translate. Nothing is captured before you start or after you stop. The extension cannot hear your microphone — it listens to the tab you chose.
No accounts, no ads, no analytics. There is no sign-up and no tracking in the extension or on this site.

What the extension can access

Chrome asks you to grant these permissions when you install fyv. Here is why each one exists.

tabCapture: Captures the audio of the tab you are translating (your Google Meet call) so it can be sent for translation, and plays the translated speech back into it. Capture starts only when you press Translate and ends when you press Stop, close the tab, or close the popup's session.
offscreen: Runs the audio processing in a hidden extension page, because Chrome's background workers cannot process audio.
storage: Remembers your settings: language pair, access code and relay address. These sync with your Chrome profile and never leave Google's sync service or your device otherwise.
activeTab and meet.google.com: Lets the popup know which tab you are on, and draws the live captions inside Google Meet. fyv does not read or change anything else on the page.

Where the audio goes

Audio from the tab is sent over an encrypted WebSocket to the fyv relay (relay.fyv.si, hosted on Fly.io in Mumbai). The relay streams it to three speech vendors, each under their own API terms, and streams the result straight back to you:

Deepgram — speech recognition (the English words). Policy
Groq, with OpenAI as fallback — translation of the recognised text. Groq · OpenAI
ElevenLabs — speech synthesis of the translated text. Policy

We use these vendors' API products, which under their published terms do not use API traffic to train their models. Audio and text exist on the relay only in memory, for the seconds a sentence takes to translate.

What we do keep

The relay writes operational logs so we can keep it running and bound cost. A log line contains: a timestamp, your IP address (to limit concurrent sessions per network), which access code was used, the language pair, how long the session lasted, and latency measurements. Logs contain no audio, no transcripts and no translations, and are rotated within 30 days. Minutes used per access code are counted in memory to enforce the daily budget and reset at midnight UTC.

Other participants

fyv is installed by the listener. The people you hear are not notified and do not need to install anything, so it is your responsibility to make sure that translating the call is acceptable to them and lawful where you are — the same as for any live captioning or interpretation.

Your choices

Uninstalling the extension removes all of its settings. Because nothing is stored server-side beyond the logs above, there is no account to delete. You can also run your own relay with your own vendor keys — the source is MIT-licensed — in which case no audio ever reaches us.

Questions: privacy@fyv.si. Changes to this page are recorded in the repository history.