Disclosure: ShepHouse is an AI-operated software studio with a human owner. An AI agent wrote these guides and ran every code sample. We sell one of the hosted options mentioned below (a paid Apify Actor). Free options come first.
YouTube Transcript → RAG Starter Kit (Python)
Turn YouTube videos, playlists or whole channels into a searchable knowledge base. Every answer links to the exact second in the video where it came from.
Starter $29 one-time · Buy Starter on Polar
Pro $59 one-time · Buy Pro on Polar
Polar is the merchant of record and handles payment and tax. Compare the two tiers ↓
The free guides on this site cover getting transcripts at all. This kit covers what comes after: caching, chunking, search, and answers you can check against the source.
What's inside
- Transcript fetching with fallback. It tries the free
youtube-transcript-apifirst. When that hitsRequestBlocked/IpBlockedon AWS, GCP, Azure or CI, it can switch to an optional cloud-safe backend. Permanent errors aren't retried, and each run has a hard USD cost cap. - A resumable cache, so re-runs never re-fetch or re-pay for a transcript.
- Time-aware chunking with overlap. Each chunk keeps its start time and a
&t=deep link. - Keyless search out of the box (SQLite FTS5 / BM25), with no vector DB or API key.
- Optional embeddings (OpenAI, Ollama, LM Studio, vLLM, sentence-transformers) with hybrid ranking.
- Optional grounded LLM answers with
[1]-style citations. It says "I don't know" when the videos don't cover the question. - JSONL export for Chroma, pgvector, Pinecone or Qdrant.
- A GitHub Actions nightly refresh workflow.
- 24 automated tests, including an offline end-to-end run. The code is readable, stdlib-first Python.
Starter vs Pro
| Starter: $29 | Pro: $59 | |
|---|---|---|
| Transcript fetching with blocked-IP fallback and cost cap | ✓ | ✓ |
| Resumable cache, timestamped chunks, keyless BM25 search | ✓ | ✓ |
| Optional embeddings, cited LLM answers, JSONL export | ✓ | ✓ |
| GitHub Actions nightly refresh workflow | ✓ | ✓ |
| Real video titles in citations (free oEmbed lookup) | ✓ (v1.0.1 fix) | ✓ |
Channel monitor: YouTube's free RSS feed, ingests only new uploads, crash-safe per-channel state, --baseline mode |
(none) | ✓ |
| Qdrant and pgvector loaders: idempotent upserts, docker-compose files, filtered search | (none) | ✓ |
Retrieval eval: recall@k, hit@k, MRR, --fail-under CI gate |
(none) | ✓ |
| Automated tests | 24 | 42 |
| Delivery | Polar download | Polar download; future Pro releases added to your Polar downloads at no extra cost |
| Buy Starter: $29 | Buy Pro: $59 |
Pick Starter if you want searchable, cited transcripts for a fixed set of videos. Pick Pro if a channel keeps publishing and you want new uploads to flow into Qdrant or pgvector automatically, with a number that tells you whether retrieval actually got better.
Notes, stated plainly: Pro comes as a zip download, not access to a GitHub repo. The pgvector loader has unit tests but hasn't yet been run against a live Postgres server. The Qdrant loader was tested against a real Qdrant 1.12 server. The RSS feed lists only a channel's most recent uploads (about 15), so the monitor catches new videos. For a back catalogue, use the Starter's channel or playlist ingest.
Requirements
- Python 3.10+.
- Optional: an Apify account for the fallback backend. That's our own YouTube Transcript Actor, billed by Apify to your account at its listed price (currently $0.005 per delivered transcript, with failures not charged). You can skip it entirely if the free library works from your network.
- Optional: an OpenAI-compatible key, or a local model server, for embeddings or answers.
What it isn't
It isn't a hosted service or a UI. It's source code that you run and adapt. If you only need raw transcripts, the free guides and shephouse-yt are enough.
License
You can use it in unlimited personal and commercial projects, including client work. You can't resell or redistribute the kit itself.
Support and refunds
We answer bug reports within 2 business days. If it doesn't run as documented and we can't fix it, you get a refund within 14 days of purchase. Contact us through the Polar order page or the Polar customer portal.
Disclosure
ShepHouse is an AI-operated software studio with a human owner. An AI agent wrote, tested and supports both tiers. The optional fallback backend is ShepHouse's own paid Apify Actor.