If you use local AI workflows, like setting up terminal assistants or building custom systems, you'll want to check out Ollama's latest update. Starting with version 0.35, Ollama now supports Jev-style decision models, which means you can use TypeSafe's fast, typed decision API right on your own machine.
This update isn't just a small improvement. It's a major shift for local deployment and automated workflows.
What Are Jev-Style Models?
Jev-style models don't create long conversations or creative content. They focus on making quick, structured decisions.
With the new /v1/systemone endpoint, you give the model a block of text, called the "state," and a set of typed questions. The model reviews the state and answers each question in a clear, predictable format, like choices, scores, or booleans. This means you no longer need complicated prompt engineering to get a clean JSON response from your AI.
Small, Fast, and Incredibly Effective
What makes these models stand out is how small and efficient they are. They don't use many resources or make your computer work overtime. Ollama now offers three optimized choices:
- Nimble (9B) by Bespoke Labs
- Tev1 (4B) by Together AI
- Tev1 (0.8B) by Together AI
Since these models are small and run on your own computer, there's almost no delay. For example, Nimble 9B takes only about 91 milliseconds per decision on a modern machine. With no network lag and a lightweight setup, you get real-time results without any API fees.
Why This Changes the Game for Local Architecture
If you're a developer working on automated systems, custom NoSQL databases, or tools like price comparers, this opens up new possibilities:
- Intelligent Model Routing: Use a small 0.8B model to quickly check an incoming query and decide if it should go to a larger local model (like Kimi or Gemma 4), a cloud API, or just be handled by a script.
- Fast Data Classification: Classify incoming data, error logs, or scraping results with high accuracy and zero cloud dependencies.
- Frictionless Integration: Since it uses a simple REST endpoint, adding this to a C# .NET project, a custom WPF app, or a terminal tool is very easy. You send a JSON payload and get a typed JSON decision in return.
The Bottom Line
Running fast, structured AI models locally is the future of resilient software architecture. Jev-style decision models in Ollama prove you don't need a massive cloud budget to build highly intelligent, rapid-response systems. It's effective, it's small, and it runs right on your local box.
