As an administrator, open Operator → Publish pipeline → Recording media and publication. First choose what stays on the server after processing, then choose what people receive. Save applies these choices to newly accepted recordings; existing and active recordings keep their saved policy.

Choose what to keep

Keep after each recording	What happens after processing	Rerun transcription?
Nothing	Delete recording media, including audio, video and raw packet logs. Keep transcription artifacts, including published JSON.	No
Full audio + video	Keep captured audio, available camera video and associated recording files under Storage policies.	While the source exists
Audio-only	Keep captured audio and associated recording files under Storage policies, without camera video.	While the source exists

Nothing captures audio temporarily for one processing run. It deletes source media after publication succeeds or processing permanently fails. Temporary retries within that run remain possible. If transcription permanently fails, there may be no usable transcript. Transcripts, meeting information and diagnostic logs follow their existing retention policies.

Temporary audio -> transcription -> JSON publication -> delete recording media
                         |
                   terminal failure -----------------> delete recording media

This removes Cassini-managed recording media; it does not mean audio never touches disk. External backups and snapshots are outside this policy.

Choose what to publish

• Include audio (.opus): playable audio with the transcript and meeting information. Captured video stays on the server; it is not published.
• Transcription only (.json): transcript, speaker blocks, summaries, metadata, tags and annotations without audio playback.

Choosing Nothing selects JSON automatically and disables audio publication. Both retained-media choices allow either publication format. Publishing JSON alone does not delete retained source media. Reruns keep the original publication format.

Enable transcription before selecting Nothing

Transcription is optional and starts off. Downloading a model does not enable it. Keeping no recording media requires enabled transcription with a ready model.

1. Select Go to transcription settings in the setup notice. The button scrolls to the transcription options and moves keyboard focus there.
2. Choose a quality level. If the model is Not installed, select Download model. If it is Installed — runtime check needed, select Check model.
3. Once it is Ready, select Enable transcription or Use this model.
4. Select Save to apply the model and recording choices together.

If downloads are disabled, follow the model panel’s offline pack/import instructions. With transcription off, audio publication can still work without a model, but the Nothing option cannot be saved.

Confirm deletion

After a new recording finishes, open its job details and wait for Media deleted. A JSON meeting without playback is not proof that server cleanup has completed. If cleanup fails, the operator shows the error and retries automatically. After fixing the reported problem, Retry media deletion requests another cleanup attempt; it does not rerun transcription.

Use Operator → Storage to configure how long retained recordings, output archives, attempt history, logs and published Nextcloud meetings remain. Those age-based policies are separate from deleting recording media after processing.