How do you run System One decision models locally? (stackness.dev)

A new breakthrough in running System One decision models locally has emerged, primarily driven by Ollaya, which recently released version 0.7.3 on September 27, 2026. This runtime allows users to efficiently deploy open decision models like winnow and kev on their local machines by simply installing the runtime, importing the models, and setting the appropriate SDK configuration. Ollaya stands out by supporting a broader range of model families behind a unified API, with impressive response times ranging from 8 to 89 milliseconds for various models, significantly improving upon hosted alternatives like Jev, which has longer latency. This development is significant for the AI/ML community as it offers a local solution that enhances privacy, reduces network latency, and increases control over model deployment. The innovative runtime architecture distinguishes between the model's weights and the runtime environment—allowing variations in output depending on the chosen runtime configuration. This flexibility is essential as even minor differences in runtime can lead to discrepancies in model predictions. The Ollaya framework, supported by additional libraries like laya-mlx for Apple Silicon, presents a user-friendly approach for leveraging AI decision-making capabilities without relying on external APIs, making it a valuable tool for developers looking to prototype and fine-tune AI applications efficiently.