Upload audio or recordings (.wav, .mp3, .m4a, .mkv — e.g. OBS), run transcription and summary on the server, then download or copy results.
Outputs live under your pipeline root (often this repo or ~/meeting-pipeline): results/<stem>.transcript.txt, .transcript.json, .summary.txt, optional .speakers.json. Audio moves to processed/. Open Review on a finished file for the full view.
Edit Markdown (or .txt) templates. Saves go to
— and override bundled files in the repo. Optional alias file:
—
“Delete my copy” only removes the file under your prompts folder (not the repo original). After deletion, the pipeline uses the bundled template again until you save an override.
Each template file becomes a preset named like its basename (e.g. kurz.md → preset kurz). Templates should include {{TRANSCRIPT}}.
Check presets in the order you want: first = primary file (*.summary.txt), others = extra files (*.summary.<name>.txt). Uncheck to remove.
Matches MEETING_LANGUAGE for pure DE/EN; Auto and Mixed leave transcription unlocked so Whisper can follow code-switching. Mixed adds a short note for the summarizer (your templates already allow DE/EN).
From /api/tags on this server. CLI/cron still use OLLAMA_MODEL unless you use the web UI.