The Recap Now Delivers Itself

The Discord bot can now post the full session transcript and AI recap to a text channel of your choice, as soon as transcription finishes. A user asked for it last week. It shipped today.

What gets posted

Record a session as usual, with /lp or an auto-record rule. When transcription completes, the bot posts to the channel you picked:

• An embed with the session name, campaign, duration, and the two-sentence recap.
• A session-notes.md file: the structured notes (summary, key events, NPCs, locations, combats, loot, open threads) followed by the full speaker-labelled transcript.
• A link to the session on LorePanic.

The transcript is a file attachment because a three-hour session is far past Discord's message limits.

Turning it on

• Per session: the /lp wizard's confirmation screen has a channel picker, "Post transcript to a channel (optional)".
• For auto-record: /lp-autorecord asks for a transcript channel after the recording mode. Set it once; every session that rule records gets posted.

It's opt-in: pick no channel and nothing is posted. When posting is enabled, the consent message at recording start says where the transcript will go. The bot verifies it has View Channel, Send Messages, and Attach Files when you pick the channel, not when the session ends.

Delivery is restart-safe: the pending post lives in the database, not in the bot's memory, so a redeploy mid-transcription doesn't lose it. If the channel was deleted or permissions revoked, the bot records the reason and stops retrying. The transcript stays on LorePanic either way.

The diarization bug we fixed on the way

Testing on a real two-person call, the transcript came out as two monologues: everything one speaker said in a single block, then everything the other said. The merge logic that interleaves speakers by timestamp was correct; the problem was that we never asked Voxtral for segment timestamps, so each track came back as one undated blob and there was nothing to interleave. Adding timestamp_granularities to the request fixed it. Transcripts now alternate speaker by speaker, each line named, which is the per-speaker transcript we promised in the April post.

What's next

The same user also asked for API access to download sessions programmatically. If you'd use that, tell us.