Ollaya ships local, open alternative to TypeSafe's Jev
The open-source tool serves TypeSafe's own API from a local server and, on its own tests, matches Jev's accuracy at a fraction of the latency.
Ollaya, a new open-source tool for running "decision models" locally, reached the Hacker News front page Friday with 108 points and 32 comments. It serves TypeSafe's own Jev request and response format from a local HTTP server, so the official TypeSafe Python SDK works against it unchanged, the project's site says.
The project packages five open model families under an Apache-2.0 runtime, including Laya from Convai Innovations and a Qwen3.5-based model called Decider. It runs them through ONNX Runtime on a local CPU or GPU. Weights are pulled from their authors' own Hugging Face repositories and checked against a sha256 hash, the site says.
On its own benchmark, Ollaya put a five-question request to its Laya model at 8 to 10 milliseconds on an Nvidia RTX 4090. That compares with 236 to 276 milliseconds for TypeSafe's hosted Jev API in third-party tests it cited. It also said Laya's calibration error is 0.081, versus 0.246 for Jev, after temperature fitting. Nobody outside the project has verified either figure.
Ollaya is the third open, local alternative to TypeSafe's paid Jev service to surface online this week, after developer Dmitri Sotnikov's Lev and AlexWortega's openjev model on Hugging Face. Unlike those single-model projects, Ollaya ships as a desktop app and command-line tool for macOS, Windows and Linux, plus a Docker image, with more models planned.