Jev Model

Training code for SearchJev, a fast calibrated System-1 model for search agents: given a search state and a decision schema, it returns typed predictions over legal field values in one forward pass of a Qwen3.5 backbone.

A decision model from Command Code in two sizes, Agr and Agr-flash: send a state as text or JSON with typed questions and get a probability for every option, with all questions about the same state answered in one forward pass.

A fast, single-pass 2B System 1 model for robots' physical judgment: it reads a short text of what a robot senses and returns probabilities for the questions that matter in the moment, scoring 0.81 on its own benchmark against TypeSafe Jev's 0.28.

A self-trainable 2B browser-agent decision model that speaks TypeSafe's /v1/systemone protocol, so an existing jev-ultrafast agent is repointed with a one-line base-URL change. Runs as a 1.6 GB Q4_K_M GGUF under llama.cpp at about 236 ms per decision.