YouTube Transcript API + Summary & RAG

Extract captions from public YouTube videos and return timestamped JSON, plain text, SRT, WebVTT, and deterministic intelligence artifacts. No YouTube API key, browser, Whisper, or external LLM is required.

What it returns

One dataset item per input URL:

• normalized YouTube URL and video ID
• caption language and whether captions were auto-generated
• timestamped JSON segments
• optional plain text, SRT, and WebVTT
• stable diagnostics for unavailable videos, disabled/missing captions, translation failures, and blocked requests
• optional intelligence pack:
  • multi-sentence extractive summary
  • timeline-wide chapters with evidence
  • explicit action items with owner, due hint, priority, and timestamp
  • selective notable claims with citations
  • keywords
  • segment-aligned RAG chunks

The intelligence pack is deterministic. It does not invent facts or send transcript data to an AI provider.

Input example

{
  "urls": [
    "https://www.youtube.com/watch?v=aircAruvnKk"
  ],
  "languages": ["en", "es"],
  "analysis_mode": "intelligence",
  "formats": ["json", "text", "srt", "vtt"],
  "proxy_mode": "auto",
  "chunk_chars": 1200
}

Supported URL forms include youtube.com/watch?v=, youtu.be/, /shorts/, and /embed/.

Proxy strategy

• none: direct access only.
• auto: direct first; retry through Apify Residential Proxy only when YouTube reports IP blocking.
• residential: use Apify Residential Proxy from the first request.

Blocked residential exits are retried up to three times with a fresh HTTP client. Missing captions, private/unavailable videos, and other terminal errors are never retried.

A proxy is not retried for videos that are private, unavailable, age-restricted, or genuinely have no captions.

Pay-per-event

• transcript-extracted: one event per successfully extracted video.
• intelligence-pack: one additional event per successful video in intelligence mode.
• Invalid URLs and failed extractions are not charged.

Limits

• 1–100 URLs per run.
• Caption-first V1: no audio download or speech-to-text fallback.
• Uses YouTube's undocumented caption endpoint through youtube-transcript-api; upstream changes can require maintenance.

Local verification

uv run --with-requirements requirements.txt --with pytest python -m pytest tests -q
uv run --with-requirements requirements.txt python -m src.main \
  --input samples/input.json \
  --output /tmp/youtube-transcript-output.json
apify validate-schema