Subtitle RAG Chunker
Pricing
$5.00 / 1,000 document with useful outputs
Subtitle RAG Chunker
Convert SRT and WebVTT subtitles into timestamped JSON chunks for video transcript RAG, semantic search and vector databases. Preserve cue timestamps, source provenance, stable chunk IDs and SHA-256 hashes.
Pricing
$5.00 / 1,000 document with useful outputs
Rating
0.0
(0)
Developer
Dakota Myers
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
4 days ago
Last modified
Categories
Share
Give me valid SRT or WebVTT subtitle text, or a public HTTP(S) text URL, and I reliably return deterministic timestamp-aware RAG chunks without splitting a cue. Records carry source and document identity, cue numbers where supplied, first/last cue timestamps, hashes, chunk order, and chunking configuration.
Submit one or more subtitle sources through the items array. rawText takes precedence over sourceUrl or fileUrl. chunkSize and overlap are deterministic parameters that materially affect output.
Output and repeatability
The default dataset contains cue-preserving chunks with stable documentId, chunkId, SHA-256 hashes, timestamps, cue metadata, source provenance, and chunking parameters. Identical subtitle text and semantic parameters produce the same useful records and stable IDs.
The default key-value store record SUMMARY contains aggregate run status, submitted-document count, total chunk count, per-document outcomes, and a UTC processedAt timestamp. processedAt is volatile metadata and does not participate in stable chunk identity.
A failed input is isolated from successful siblings rather than discarding useful batch output.
Limitations: tested for ordinary SRT and WebVTT timing syntax, multiline cues, simple tags, duplicates, gaps, and malformed timestamps. It does not support TTML, ASS/SSA, subtitle OCR, translations, speaker diarization, or live streams.
Billing and limits
One document-converted event is charged only after a document produces supported output saved to the dataset. Empty or failed documents are not charged. Platform costs are included; no start or automatic dataset-item charge. Up to 25 documents, 2 MB UTF-8 text per document, and a 120-second run timeout. The Actor stops at the spending limit. Public text downloads must be accessible without credentials.
