Positioning The local model runtime Ollama supports Jev-style decision models as of version 0.35, serving typed decisions on the local machine through a /v1/systemone endpoint, announced on 2026-09-29.
What it does A request sends a text state with a set of named questions, and a model running on the machine answers all of them in one request, returning answers in the same structure as Jev: choice with per-option probabilities and confidence, noul as the probability of yes, and score as the probability-weighted value with a legend. The response carries the model name, answers keyed by question id, and token usage, matching the fields Jev clients expect. The official TypeSafe Python SDK works by pointing TYPESAFE_BASE_URL at the local port and setting TYPESAFE_API_KEY to ollama; curl against http://localhost:11434/v1/systemone works as well. Use cases named in the official blog post include ticket triage, model routing, and content or safety moderation.
Characteristics Three decision models ship initially — nimble, tev1, and tev1:0.8b — fetched with a single ollama pull nimble command, with more models including Ollama-cloud-served ones planned. Running locally removes the network round trip, and requests stay on the machine, so state does not travel over a network. In the vendor's own example, Nimble 9B averaged 91ms per decision on an M5 Max (vendor-reported).
When to use Suits running a System One-compatible decision service on a personal machine with the least setup, especially for prototyping and latency-sensitive real-time decisions; the endpoint requires version 0.35 or newer. The model selection is still small, and tasks that depend on probability quality should validate thresholds on labeled data before adoption.