Ollaya runs open decision models on your own hardware, the way Ollama runs LLMs.
Typed questions in, calibrated answers out, in milliseconds: routing, guardrails, triage and tool approval without an LLM call.
Website · Models · Bench · Docs · Releases · Hugging Face
curl -fsSL https://ollaya.dev/install.sh | sh
ollaya run winnow:e4b --preset triage "I was charged twice this month and want a refund."29 model families, from millisecond encoders that run on a CPU (laya, nli, gliclass, von, decima) to
decoders that take a GPU (winnow, pplx-decider, clef, jeeves, nimble, kev, jeb, d3 and more), behind a
self-hosted, Jev-compatible API. Their accuracy, calibration and speed on our own GPUs and CPUs, with the raw data,
are at bench.ollaya.dev.
Weights always come from their authors' own Hugging Face repositories, pinned to a commit and verified by sha256. Ollaya never re-hosts them.
Created and maintained by Mert Cobanov (@mertcobanov). Apache-2.0. Ollaya is an independent project, not affiliated with Ollama or TypeSafe.
