castrag
github.com/snowmac/castrag- Category
- Developer Tools
- Rank
- No. 4121Tools index
Previous survey · No. 3887 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Data, Retrieval & Knowledge
- GitHub
- 1 stars
- Date
About
Castrag is a resumable pipeline that transcribes large podcast or audio archives and turns them into structured per-episode summaries, a semantic search index, and a retrieval-augmented Q&A system grounded in the source material, scripted around the OpenAI API. It was built to process 230 episodes (about 145 hours of audio) for roughly $33, making a large audio archive searchable and answerable instead of requiring the full transcript to be read or fed into an LLM's context window.
At a glance
castrag turns an audio archive into searchable transcript passages and answers that cite episode files. It is worth trying if you repeatedly need to find what someone said across recordings you have permission to process.
Good fit
People who want a scriptable research workflow for their own recordings and can inspect the transcripts.
Consider something else when
Choose another workflow if audio and excerpts must stay off external services or you need timestamped citations.
Start here
Index one short recording you own. Ask a question with a known answer, then check the cited transcript for the wording and surrounding context.
Cost and requirements
MIT-licensed Python scripts with FFmpeg and an OpenAI API key. Transcription, embeddings and answer generation use the API.
Limits to know
Answer generation sends retrieved excerpts to OpenAI. The script asks the model to cite transcript files. Check the original audio before relying on a quote.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Shows a working, cost-transparent architecture (transcription + embeddings + grounded RAG) for making huge audio archives searchable and queryable, a pattern reusable well beyond podcasts.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.