Tried Jev when it came out last week and it's crazy fast. Heaps of the LLM calls in my agents were already structured data anyway, so I thought I'd try moving them over.
It worked, so I wrapped it in a skill that does it end-to-end. Tested on popular repo and it all worked, see the demo in GitHub repo.
You can just run /jevify in your repo. It marks which ones could move to Jev, and converts to use Jev (TypeSafe, OpenRouter, or Vercel).
The scan works without a Jev key if you just want to see what it finds.
Not affiliated with TypeSafe. Would love to hear how it goes!
Give it a try! Github: https://github.com/Kevin-Zhouu/jevify/
Comments URL: [https://news.ycombinator.com/item?id=49812519](https://news.ycombinator.com/item?id=49812519)
Points: 2
source & further reading
news.ycombinator.com — original article
Ask HN: Engineers on highly polished products, how are you using agentic coding?
Show HN: Making agent decisions better and traceable
Ask HN: When is fine-tuning a small LLM worth it?