Podhead / Pipeline

Cosine similarity cutoff used when comparing new insights against existing ones. Lower values catch more near-duplicates; higher values keep more as distinct.

Limit existing SQLite episodes processed by the Transcription Queue and Manual Podhead Pipeline. Leave blank to process all eligible statuses.

Process SQLite episodes that already have a transcript or a local source file and still need summary and insights.

Fetches NEW YouTube videos from every source whose status is exactly Tuning, then transcribes and extracts insights. Videos already saved are skipped. Videos shorter than 10 minutes are saved with status "Too Short" and not transcribed. Uses the Extraction prompt selected at the top of the page.

Process manual Podhead episodes from SQLite. Uses transcript text, a local source video path, or a YouTube video ID, then saves results back into SQLite.

Re-run transcription on episodes saved with status Transcription Failed. Each one gets 3 fresh Supadata attempts. Successful ones are fully processed and saved as Processed; the rest stay Transcription Failed so you can retry again later.