| name | youtube-ingest |
| description | Ingest YouTube video transcripts into the transcript database using Gemini API |
| tags | ["youtube","transcript","ingest","gemini","research","external"] |
| version | 1.0.0 |
/youtube-ingest — YouTube Transcript Ingestion
Extracts transcripts from YouTube videos via the Gemini API (native YouTube processing) and stores them in the transcript SQLite database.
When to use
- User or Captain requests transcript extraction from YouTube videos
- Research needs to ingest external video content for analysis
- Building knowledge base from video sources
Command Grammar
/youtube-ingest <url> — Ingest single video
/youtube-ingest <url1> <url2> ... — Ingest multiple videos
/youtube-ingest --channel <handle> --days <N> — Ingest recent videos from channel
Execution
Run the handler script:
bash ~/.openclaw/scripts/youtube-ingest.sh "$@"
How it works
- Takes YouTube URL(s) or channel handle
- Calls Gemini API (
gemini-3-flash-preview) with native YouTube video processing
- Extracts: transcript (complete), description, upload date
- Stores in
/root/.openclaw/transcripts.db with metadata
- All entries marked: source=external, trust_level=0.9, trusted_by=Robert
Database Schema
videos(video_id, title, publish_date, url, description, transcript,
summary, key_insights, source, trust_level, trusted_by, channel, created_at)
Notes
- Gemini API processes YouTube URLs natively — no cookies, no proxy needed
- ~30s per video, 20-40K chars of transcript each
- Rate limit: 2s between API calls to avoid 429s
- YouTube blocks direct access from this VPS (cloud IP) — ONLY the Gemini API method works
- Cost: Gemini Flash pricing per video (~0.5M tokens input per video)
Known Issues
- Gemini sometimes returns wrong upload dates — use channel posting cadence for verification
- Very long videos (>2hr) may timeout at 180s — retry with longer timeout
- See chart: issue-youtube-cloud-ip-block
Intent: Informed [I18]. Purpose: External knowledge ingestion.
