The Proxy Company
@TheProxyCompany
Building the infrastructure for private, personally-owned AI.
The future of AI will not be metered. It will be owned by you. Our work continues in Brooklyn.
I see you never tried @TheProxyCompany 's proxy-structuring-engine
Everybody knows how much I hate forced, guided output libraries! 🤬 Sure, you get structure, but at the cost of true content quality. This is the way: Let your big model speak freely, THEN just use a tiny 0.6B model to structure the answer. Don’t cage the genius!
If you haven't seen, @TheProxyCompany is doing this with PSE as well as an engine that bolts into various inferencing setups. @jckwind's been working at it for quite a while, been a little busy to dig in recently but worth checking both out.
Don't use structured output mode for reasoning tasks. We’re open sourcing Osmosis-Structure-0.6B: an extremely small model that can turn any unstructured data into any format (e.g. JSON schema). Use it with any model - download and blog below!
Strong validation from @secemp9! This level of reliability for structured output is exactly why we engineered the Proxy Structuring Engine – tackling the core unpredictability of AI head-on. Explore the tech: proxy.ing/pse
this is pretty much SOTA for structured output for llm. This is better than anything that is on arxiv atm. I'm not kidding
this is pretty much SOTA for structured output for llm. This is better than anything that is on arxiv atm. I'm not kidding
1. LLMs can't help but introduce random errors. Hallucinations, broken structures, and unpredictable actions aren't acceptable for dependable systems. We're sharing the foundations for guaranteed AI control and reliable agentic execution. It's time to build AI you can trust.