How VelvetSpeak works, honestly.

VelvetSpeak listens to your conversation on Even Realities G2 glasses. It uses AI to suggest what you could say, one card at a time.

Suggestions can be wrong. The wearer decides what to say. The glasses never speak for you.

Live audio is streamed for transcription, processed, and then discarded. It is used to produce the next card, not to build a record.

Usage records keep durations and counts, never the words. Memory capture is controlled by the wearer in Settings, and it can be wiped at any time.

VoiceLock is optional and off by default. If you set it up, it uses short samples of your own voice to build a recognition profile that tells your voice apart from other people in the room.

We keep that profile, not your recordings. The raw enrollment audio is discarded once the profile is built. You can delete the profile at any time in Settings or with Wipe all.

Recording and transcription laws differ by region. Many places require everyone in a conversation to agree before it is recorded or transcribed.

Beta testers are responsible for getting that agreement where it is required. When in doubt, tell people about the glasses before you start.

The people you talk to deserve to know. VelvetSpeak does not announce itself, so telling people an AI is helping is the wearer's responsibility, the same as getting their agreement to be transcribed where the law requires it.

VelvetSpeak is not a safety tool. If a conversation involves danger to anyone, stop using the glasses and get real help.

This is a small closed beta. Things will be imperfect. Access can be paused or revoked.

If anything feels uncomfortable, report it to the person who invited you. They can reach the team directly.