AI summarization app
TLDWatch
Paste a YouTube link, long text or a file and get a summary you can read or listen to, in about a minute.

What makes it different
An LLM that watches the video
Gemini reads the video, PDF, image or text and writes three summary lengths with key points, timestamps or page references, and a verdict on whether the original is worth your time.
- Gemini API
- YouTube Data API
Listen instead of read
Summaries can be read aloud by the device's own voices, or by natural AI voices for signed-in users.
- ElevenLabs
- Web Speech API
Rules that don't depend on any vendor
Limits, ownership and file checks live in a framework-free core. Every outside service sits behind an interface, so the tests run against in-memory fakes and never call an API.
- Ports and adapters
- Vitest
Costs kept in check
A YouTube summary is made once and shared by link. Daily limits apply per guest, per network and per member, with a global cap on video minutes and voice characters.
- Supabase Postgres
- Signed cookies
Uploads handled carefully
Files are checked by their contents in the browser and again on the server, stored privately only while they're summarized, then deleted. A daily job cleans up anything left behind.
- Supabase Storage
- Vercel Cron
Checked on every push
GitHub Actions runs the type check, lint and tests on every pull request and every push to main.
- GitHub Actions
- Next.js
- Supabase
- LLM APIs
- Text-to-speech