Case studytranscriptyt
Transcripts, on request.
transcriptyt is a REST API for YouTube transcripts and for video and channel metadata, with API keys and rate limits by plan.
- Status
- Live at transcriptyt.com. It also feeds sponsoredyt.
- Role
- The API, its caching and its plans.
- Built with
- Python
- FastAPI
- PostgreSQL
- Clerk
- Sentry
- Docker
The problem
Getting YouTube transcripts in code means dealing with blocked requests, retries and caption formats. transcriptyt puts all of that behind one authenticated API.
What was built
- Transcripts in JSON, plain text, SRT or WebVTT, with a language order of preference and a filter for manual or auto-generated captions.
- Video metadata, including the title, channel, length, views, publish date and members-only status, and channel metadata.
- API keys and plans through Clerk, with per-minute and monthly limits, and rate-limit headers on authenticated responses.
- A public health check that reports the database and the circuit breaker.
How it works
- FastAPI with a cache in PostgreSQL (JSONB): check the cache, then YouTube, then store the answer.
- A circuit breaker and retries with backoff around YouTube calls, and a fallback to the last cached answer, so responses keep flowing when YouTube pushes back.
- Security headers, a limit on request size, sanitised logs, and monitoring with Sentry.