YouTube Transcript to RAG Dataset

Extract YouTube video transcripts and convert them into structured, RAG-ready chunks with timestamps, metadata, and content hashes. No API key required.

What does this Actor do?

This Actor takes YouTube video URLs, fetches their transcripts using the YouTube transcript API, and splits them into chunks optimized for Retrieval-Augmented Generation (RAG) pipelines. Each chunk includes precise timestamps, stable IDs, content hashes, and token estimates. It also outputs the raw transcript alongside the chunks.

Why use it?

• Build knowledge bases from YouTube content without manual transcription
• Feed video transcripts into vector databases, search systems, or LLM context windows
• Get stable, deterministic chunk IDs so re-runs produce the same identifiers
• No YouTube Data API key needed for basic metadata (uses oEmbed)
• Choose from three chunking strategies to fit your embedding model's context window

Data Fields

Chunk Records

Field	Type	Description
type	string	Always "chunk"
chunk_id	string	Stable UUID based on video ID and start time
video_id	string	YouTube video ID
video_url	string	Full YouTube URL
video_title	string	Video title from oEmbed
channel_name	string	Channel name from oEmbed
duration_seconds	float	Total video duration
language	string	Transcript language code
start_time	float	Chunk start time in seconds
end_time	float	Chunk end time in seconds
text	string	Chunk text content
source_url	string	YouTube URL with timestamp parameter
content_hash	string	SHA-256 hash of chunk text
token_estimate	int	Estimated token count (chars / 4)
chunk_index	int	Zero-based index of this chunk
total_chunks	int	Total chunks for this video
crawled_at	string	ISO timestamp of when data was fetched

Transcript Records

Field	Type	Description
type	string	Always "transcript"
video_id	string	YouTube video ID
video_url	string	Full YouTube URL
video_title	string	Video title
channel_name	string	Channel name
duration_seconds	float	Video duration
language	string	Language code
transcript_text	string	Full concatenated transcript
segments	array	List of {text, start, duration} objects
word_count	int	Total word count
token_estimate	int	Estimated token count
crawled_at	string	ISO timestamp

How to Use

1. Go to the Actor page on Apify
2. Click "Start" or configure input
3. Add YouTube video URLs to the video_urls array
4. Choose your chunk strategy (timestamp, fixed_size, or sentence)
5. Set chunk duration or token limit depending on strategy
6. Run the Actor
7. Download results from the Dataset tab as JSON, CSV, or Excel

Pricing

This Actor uses the Apify platform pricing model. Costs are based on:

• Compute: ~0.5 CU per video processed (simple API calls, no browser)
• Transcript charges: 1 platform event per video processed via Actor.charge()
• Dataset storage: Standard Apify dataset storage rates apply

A typical run processing 50 videos costs well under 1 CU total.

Input Parameters

Parameter	Type	Default	Required	Description
video_urls	array	-	Yes	YouTube video URLs
languages	array	["en"]	No	Preferred transcript languages (ISO codes)
chunk_strategy	string	"timestamp"	No	Chunking method: timestamp, fixed_size, sentence
chunk_duration_seconds	integer	60	No	Seconds per chunk (timestamp strategy)
chunk_token_limit	integer	512	No	Max tokens per chunk (fixed_size strategy)
include_raw_transcript	boolean	true	No	Include full raw transcript in output
max_videos	integer	50	No	Maximum videos to process (1-500)

Output Example

Chunk record:

{
  "type": "chunk",
  "chunk_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "video_id": "dQw4w9WgXcQ",
  "video_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "video_title": "Rick Astley - Never Gonna Give You Up",
  "channel_name": "Rick Astley",
  "duration_seconds": 214,
  "language": "en",
  "start_time": 0.0,
  "end_time": 60.5,
  "text": "This is an example transcript chunk.",
  "source_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ&t=0",
  "content_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
  "token_estimate": 150,
  "chunk_index": 0,
  "total_chunks": 4,
  "crawled_at": "2026-07-04T12:00:00+00:00"
}

Transcript record:

{
  "type": "transcript",
  "video_id": "dQw4w9WgXcQ",
  "video_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "video_title": "Rick Astley - Never Gonna Give You Up",
  "channel_name": "Rick Astley",
  "duration_seconds": 214,
  "language": "en",
  "transcript_text": "This is an example transcript.",
  "segments": [{"text": "This is an example transcript.", "start": 0.0, "duration": 2.5}],
  "word_count": 1500,
  "token_estimate": 375,
  "crawled_at": "2026-07-04T12:00:00+00:00"
}

FAQ

Q: Does this require a YouTube API key? A: No. Transcripts are fetched via the youtube-transcript-api library and metadata comes from the oEmbed endpoint, both of which work without API keys.

Q: What languages are supported? A: Any language that has subtitles/captions on the YouTube video. Set the languages parameter with ISO codes in order of preference.

Q: How are chunk IDs generated? A: Chunk IDs are deterministic UUID5 values based on the video ID, start time, and chunk index. Re-running the same video produces the same chunk IDs.

Q: Can I process playlists? A: No. This Actor accepts individual video URLs only. Playlist expansion requires the YouTube Data API which needs an API key.

Q: What happens if a video has no transcript? A: The Actor logs a warning and skips that video. Processing continues with the remaining URLs.

Disclaimer

This Actor relies on the youtube-transcript-api library and YouTube's oEmbed endpoint. YouTube may change access behavior or rate limit requests at any time. This Actor is not affiliated with YouTube or Google. Use responsibly and respect YouTube's Terms of Service.