Timbre · private beta

Meeting notes that know who said what.

Timbre transcribes the call in your browser tab and labels each speaker. No bot joins the call, and the host doesn't have to approve anything. In private mode, the meeting is transcribed, labelled and summarised on your own hardware.

No meeting botUp to 8 speakers40+ languages
Timbre in 25 seconds (sample meeting)

How it works

Captures the tab, not the room.

01

Your tab and your mic

Timbre records the call audio in your browser tab plus your microphone. Your own words are labelled "You".

02

Labels up to 8 speakers

Each voice gets its own label. On Google Meet, names are suggested from the participant list and you confirm them with one click.

03

Live, then cleaned up

The transcript appears a few seconds after people speak. When you stop, a final pass tidies the speaker labels.

Full support for Google Meet. The Zoom web client works with generic speaker labels. Transcription covers 40+ languages, English tested first.

Private mode

The meeting stays on your hardware.

In private mode, Timbre refuses to start unless transcription, speaker labels and the notes model all run on your own machine or your own network. It is enforced by configuration, not by a missing key, and the model libraries run offline.

Transcription, local

Speech-to-text runs on your own GPU.

Speaker labels, local

Who-spoke-when runs as a service on your machine or another machine on your network.

Notes, local

Summaries and action items come from a local model and never leave the machine.

What we measured: across a full session in private mode, the backend and the speaker-label service talked only to the local machine.

Private beta

Join the private beta.

We are onboarding a small group of teams who want meeting notes without sending their meetings anywhere. Tell us how you meet and what hardware you have, and we will reply within two working days.

Today: Google Meet fully, the Zoom web client with generic speaker labels. Private mode runs on a machine with an NVIDIA GPU; a Mac build is planned.

You are responsible for telling people you are transcribing. Timbre shows a recording indicator and offers a consent announcement.