Meeting transcription that runs on your own hardware

The media server, the transcription and the audio all stay on your host. The whisper sidecar makes no outbound call, and the model is delivered with the deployment.

Gather can now run its whole capture path inside your own network.

The usual way to get a transcript is to invite a bot to the meeting. The bot joins, the audio goes to somebody else's cloud, and a transcript comes back. For a lot of teams that is the end of the conversation, because the rule they are working under was written long before meeting bots existed and it does not care how good the vendor is.

So the on-prem build keeps the whole path local. The media server that carries the call, the transcription, and the recorded audio all sit on your host. The transcription sidecar is whisper.cpp and it makes no outbound call at all, which is a property of the component rather than a setting you have to remember to switch on.

The model file is delivered with the deployment and never downloaded. We publish the checksum of the exact file the feature was benchmarked with, so you can verify your copy before you trust it rather than after. The container refuses to start without a model rather than quietly falling back to something else.

Recorded audio has a retention window, thirty days by default, and a pruner that enforces it. The transcript is the artefact worth keeping and nothing touches it. A track older than the window was transcribed long ago, so deleting it loses nothing the transcript does not already hold.

If your meetings are the reason you cannot use a meeting tool, this is the build to ask about.