ollaya-dev/ollaya is a Rust API server that serves open decision models locally or in self-hosted environments. Like Ollama, it pulls and runs models by name, and it supports TypeSafe's /v1/systemone format. The latest release is v0.18.0. The provided record does not list anything as newly changed in this version. The project description also covers a CLI, a desktop app, Docker, and an MCP server.
Decision models do not generate text. Instead, they take a state such as a message, email, ticket, or JSON as input and answer choice, score, and noul (no response) style questions with calibrated probabilities. They are models for tasks that require judgment, such as routing or classification. The project offers 29 model families — including Laya, pplx-decider, Clef, Winnow, and coba — behind a single interface. The TypeSafe SDK can be connected to simply by changing the base URL, which also suggests a path for reusing existing clients.
In the self-measurement recorded in the README, the type classification accuracy of winnow:e4b was 0.722, while the TypeSafe model cited as a comparison baseline was 0.738. There is also a figure of 89 milliseconds to process five questions on an RTX 4090, but this is a result reported by the project. Therefore, it is difficult to judge performance on other models or in other environments from these numbers alone.