I've added schema support to my LLM Python library and a CLI tool: you can now provide a JSON schema - or a new concise schema DSL - and supported models will return data matching that format https://simonwillison.net/2025/Feb/28/llm-schemas/ - ThreadSky

About ThreadSky

simon.fedi.simonwillison.net.ap.brid.gy • 15 hours ago

I've added schema support to my LLM Python library and a CLI tool: you can now provide a JSON schema - or a new concise schema DSL - and supported models will return data matching that format https://simonwillison.net/2025/Feb/28/llm-schemas/

Comments

prem-k.mastodon.social.ap.brid.gy•15 hours ago

@simon ha! Right when I wanted my team to use the Adaptive Cards JSON schema in the response_format of Azure OpenAI I see your post explaining the concept 😄

Thanks a lot for making my job easier. 😉

If ever we get to meet IRL, I owe you a meal. 😊

jpallas.hachyderm.io.ap.brid.gy•10 hours ago

@simon My experiments with Gemini showed that the content of the output was affected by using the JSON API. I even saw a difference between using the API and appending the desired schema to the prompt. Have you encountered this?

simon.fedi.simonwillison.net.ap.brid.gy•7 hours ago

@jpallas I haven't done extensive enough testing to pick things up like that myself yet, but it doesn't surprise me at all to hear that's an issue

LLMs are weird!

miki.dragonscave.space.ap.brid.gy•15 hours ago

@simon Oh this'll be useful.

I'm doing a lot of work with structured outputs professionally, and being able to just try things out in the cli for quick prototyping and testing will be extremely useful.

This was definitely one feature I felt LLM was missing.

Thank you for such a great and […]

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply