AI processing
Phase A processors that run on top of collected posts. All are idempotent (re-running skips already-processed posts). All models run locally on the server CPU — no API keys, no per-call cost.
LLM classifier
Entity extraction
Uses the platform + limit settings from the LLM classifier above. Extracts PERSON / ORG / GPE / etc.
Whisper transcription (audio → text)
Downloads video media → ffmpeg-extracts audio → Whisper transcribes. Slow: ~5–15 sec per 30 sec clip on CPU. Use small batches.
Recent processor runs
| Started | Processor | Filter | Processed | Skipped | Status |
|---|