ruby_llm-providers-lms 0.3.0 brings RubyLLM 2.1’s judgments to LM Studio. Point RubyLLM.judge or a RubyLLM::Judge class at a local decision model and get typed answers back in tens of milliseconds: a probability that something is true, a choice from a fixed set with a confidence, or a score on an ordered scale, with no text to parse and nothing generated. Both of LM Studio 0.4.26’s decision endpoints are supported, /v1/decisions by default and /v1/systemone via lms_judgment_protocol, including image input on models with a vision adapter. Open-source decision models such as D1 Omni 600M, Kev 0.8B and Clef Flash work out of the box and now appear in list_models as judgment models. The release also adds five runnable examples covering the question types, a reusable triage judge, a model-and-endpoint sweep, image judgments, and a decision model screening an inbox ahead of a chat model, plus a new documentation site at https://madbomber.github.io/ruby_llm-providers-lms/ with measured results for each model and endpoint. It requires ruby_llm >= 2.1.1.