‹ BackHN Continuity

Thread

Clef: Open-weight decision models, and new RL fine-tuning platform

637 points · 217 comments · jasondavies

  1. croemer · · focus · HN ↗
    Is there a Jev-like model I can run on my Mac? Something like Ollama? Or what's the best way to play with it? Is there a cheap/free API service eg on OpenRouter?
    1. handfuloflight · · focus · HN ↗
      <a href="https:&#x2F;&#x2F;github.com&#x2F;NandhaKishorM&#x2F;laya" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;NandhaKishorM&#x2F;laya
      1. croemer · · focus · HN ↗
        Thanks for the tip! Set it up locally and it works!
    2. mpolichette · · focus · HN ↗
      Jumping on this, what about on-device?

      I&#x27;d love an privacy first on-device model i could use in iOS.

      1. okpatil · · focus · HN ↗
        We got you. How would you use it though ? Through Apps ? Would love to discuss.

        At0M: A 60M local Jev at 16 ms latency and 79% accuracy on Typed Decision

        <a href="https:&#x2F;&#x2F;at0m.pienomial.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;at0m.pienomial.com&#x2F; <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49920350">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49920350

    3. itsmeduncan · · focus · HN ↗

      [dead]

    4. okpatil · · focus · HN ↗
      We got you.

      At0M: A 60M local Jev at 16 ms latency and 79% accuracy on Typed Decision

      <a href="https:&#x2F;&#x2F;at0m.pienomial.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;at0m.pienomial.com&#x2F; <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49920350">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=49920350

    5. networked · · focus · HN ↗
      Ollaya (<a href="https:&#x2F;&#x2F;ollaya.dev" rel="nofollow">https:&#x2F;&#x2F;ollaya.dev) aspires to be Ollama for decision models. It can run Clef Flash, Laya, and other models. I&#x27;ve used it for local testing.

      You can list the decision models available on OpenRouter: <a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;models?output_modalities=decisions" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;models?output_modalities=decisions. There are currently nine.

      My AI harness uses Jev or a compatible model to approve shell commands that look safe without asking the user. (This is one security layer.) A quick benchmark (<a href="https:&#x2F;&#x2F;github.com&#x2F;dbohdan&#x2F;strument&#x2F;tree&#x2F;c2320ca142f77495ecdcadd3cd7a4c9b9445369f&#x2F;doc&#x2F;experiments&#x2F;2026-10-decision-models" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;dbohdan&#x2F;strument&#x2F;tree&#x2F;c2320ca142f77495ecd...) has shown that of five models on OpenRouter only Liquid D1 performs similar to Jev for this.

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.