Start the recording from Discord, watch the stems and the transcript appear on the Session page while the bot processes the audio, and retry or clean up any stem that misbehaves.
The Session went long, the cannery Scene got weird in the best way, and three hours from now you'll have forgotten the line one of your players threw out in the second half that you swore you'd reuse. That's the lost hour every GM knows. A Recording catches it, the Transcript makes it searchable, and the AI recap pulls the moments worth keeping. See Recordings, Transcripts, and recaps for how the three fit together.
You want a Recording (and the Transcript and recap that follow) for a Session you're about to play.
Confirm the players are in the voice channel for the Campaign before you start. The bot needs to join voice with them.
In the Campaign's play text channel, run /recording start. The bot joins the voice channel and starts recording one stem per speaker.
Play the Session. The bot records in the background; no further commands are needed while you're playing.
When the Session ends, run /recording stop. The bot leaves the voice channel, finalises the file, and starts processing the stems.
On the web, open the Session's detail page and click into its Recording. The stems and Transcript segments populate as they're processed; you can refresh to watch progress.
If a stem misbehaves (clipped audio, garbled Transcript, a speaker who got mis-attributed), use the stem actions menu to retry or clean it up.
Recording is Discord-driven only. There isn't a web-side "start recording" button, because the bot has to be in the voice channel and the players are already there. /recording start is the canonical entry point.
The Transcript is generated automatically from the Recording, in the background, once the audio is finalised. You don't trigger it manually.
You can edit individual Transcript segments on the Session page if a speaker was mis-attributed or a phrase was garbled. The recap reads the edited text, so a quick clean-up pass before generating the recap pays off.
A Recording belongs to one Session. The Session is the anchor; the Recording, the Transcript, and the eventual recap all hang off it.
The Recording page shows one row per speaker (one stem each). Each stem moves through states as it's processed:
Recorded. The raw audio is on the server; processing hasn't started.
Transcribing. The Transcript service is running against the stem.
Done. The stem's text is available, attributed to a speaker, and ready to read.
Errored. Something went wrong. Use the stem actions menu to retry or clean up.
The Transcript segments interleave the stems chronologically as they finalise. You don't have to wait for every stem to finish before you can start reading; the segments appear as they're ready.
Make sure the bot can join voice before the players arrive. /recording start fails silently if the voice channel hasn't been set up, and you'd rather find that out at 8:55pm than at 9:05pm.
