YouTube Transcripts for RAG
Pricing
from $5.00 / 1,000 transcripts
Turn YouTube videos into retrieval-ready text. Per video you get a full plain transcript plus token-bounded chunks (real cl100k tokens, your size and overlap) each carrying start and end timestamps, a deep link to the moment, and a deterministic chunk id for idempotent upserts. No API key, no proxy.
Record type
type
Optional
Which kind of record this row is. chunk is one retrieval chunk, ready to embed. video is one row per requested video with the transcript and what happened.
Video ID
videoId
Optional
The eleven-character YouTube video id.
Video URL
videoUrl
Optional
Canonical watch URL of the video.
Title
title
Optional
The video's title. Carried on chunk rows too, so a chunk is self-describing in a vector store.
Channel
channel
Optional
Name of the channel that published the video.
Channel ID
channelId
Optional
Video rows only. YouTube's identifier for the channel.
Language
language
Optional
Language code of the caption track that was actually used.
Auto-generated
isAutoGenerated
Optional
Whether the captions were machine-generated by YouTube rather than authored. Auto captions carry no punctuation reliability.
Mode
mode
Optional
Video rows only. Whether the run was asked for chunks or transcripts only.
Status
status
Optional
Video rows only. ok when a transcript was retrieved, error otherwise.
Error type
errorType
Optional
Video rows only. Machine-readable failure label when status is error, for example residential_proxy_required or budget_exceeded.
Error
error
Optional
Video rows only. Plain-language explanation of the failure.
Languages requested
languagesRequested
Optional
Video rows only. The language preference chain you asked for.
Languages tried
languagesTried
Optional
Video rows only. The languages actually attempted, in order.
Fallback used
languageFallbackUsed
Optional
Video rows only. True when the first choice language was unavailable and a later one was used.
Available languages
availableLanguages
Optional
Video rows only. Authored caption tracks the video offers.
Auto-generated languages
autoGeneratedLanguages
Optional
Video rows only. Machine-generated caption tracks the video offers.
Duration
videoDurationSeconds
Optional
Video rows only. Length of the video in seconds.
Views
viewCount
Optional
Video rows only. View count at the time of the run.
Retrieval source
retrievalSource
Optional
Video rows only. Which proxy lane and Innertube client returned the caption tracklist. YouTube refuses datacenter egress, so this normally starts with residential.
Caption lane
captionLane
Optional
Video rows only. Which lane downloaded the caption text itself. The cheap datacenter lane carries the bulk of the bytes.
Caption format
captionFormat
Optional
Video rows only. Wire format the captions were fetched in.
Segments
segmentCount
Optional
On video rows, how many caption segments the transcript has. On chunk rows, how many segments went into this chunk.
Chunks
chunkCount
Optional
Video rows only. How many chunks this video produced.
Transcript characters
transcriptChars
Optional
Video rows only. Character count of the full transcript.
Transcript tokens
transcriptTokens
Optional
Video rows only. Real cl100k token count of the full transcript, not an estimate.
Chunk size
chunkTokens
Optional
Video rows only. The token budget per chunk that was applied.
Chunk overlap
chunkOverlapTokens
Optional
Video rows only. Token overlap between consecutive chunks.
Residential bytes
residentialBytes
Optional
Video rows only. Measured TCP bytes spent on the residential lane for this video, counted after TLS and gzip.
Datacenter bytes
datacenterBytes
Optional
Video rows only. Measured TCP bytes spent on the cheap datacenter lane for this video.
Full transcript
fullText
Optional
Video rows only. The whole transcript as plain text.
Chunk ID
chunkId
Optional
Chunk rows only. Deterministic id of video id and content hash. Re-running produces the same id, so upserts stay idempotent.
Chunk index
chunkIndex
Optional
Chunk rows only. Zero-based position of this chunk in the video.
Chunk text
text
Optional
Chunk rows only. The chunk's text, token-bounded to your requested size. This is the field to embed.
Tokens
tokens
Optional
Chunk rows only. Real cl100k token count of this chunk.
Start
startSeconds
Optional
Chunk rows only. Where this chunk starts in the video, in seconds.
End
endSeconds
Optional
Chunk rows only. Where this chunk ends in the video, in seconds.
Deep link
timestampUrl
Optional
Chunk rows only. Watch URL that jumps straight to the moment this chunk starts, so a citation is clickable.
