GitHub - jaredpalmer/kev: tiny Jev-like family of decision models built on top of Qwen3.5 you can train and run on your own

kept by eddie sep 22

Kev provides small, trainable decision models based on Qwen3.5 that support multi‑question System One API and can be run locally.

Jev introduces a new shape of LLM—System One, aka Decision Models
The piece presents Jev, a TypeSafe AI decision‑model that turns text inputs into probabilistic scores for yes/no, choice, and rating tasks, explores its low cost, bias risks, and community hacks.
also on Kev, #open-source, #ai-coding, #developer-tools
Jev’s Architecture Unmasked — archerhume
By probing Jev's API with 10k calls, the author infers Jev is a causal transformer (possibly sparse MoE) that reads shared state and outputs decision probabilities directly without text generation.
also on typesafe, #software-engineering, #ai, #developer-tools
Jev is to tool use what RAG is to context
RAG pre-selects model input context while Jev pre-selects model output options; both constrain the model to a developer‑defined frame, contrasting with fully agentic models that decide dynamically.
also on typesafe, #ai, #developer-tools