Ollaya – Ollama for open-source, Jev-style decision models

Ollaya is an open-source decision model that delivers answers in milliseconds, running on a user's own hardware. It enables private, on-device processing of user data, ensuring the protection of sensitive information. The decision-making process is swift, requiring only about 10 milliseconds for a five-question request to Laya, thanks to the HTTP API's efficient handling.

Ollaya supports a variety of models, including English, multiple languages, language-specific fine-tuned models, and a router for choosing the most appropriate model. Future developments will include additional decision models, such as von and GGUF LLM-based models powered by llama.cpp.

The software is designed to operate securely on a user's machine, utilizing ONNX Runtime, the CPU, or an NVIDIA GPU. The server listens on the local loopback address (127.0.0.1) by default, ensuring secure, internal communication. The weights for the models are obtained directly from their original sources on Hugging Face, with each commit pinned to a specific SHA256 hash for integrity and verification purposes.

Ollaya adheres to the Apache-2.0 license, allowing for free use, modification, and distribution of the software. It is compatible with various platforms, including macOS, Windows, and Linux, and can be easily deployed using Docker images. The software remains completely self-contained, requiring no additional meters or billing for usage, and provides probabilistic outputs that can be evaluated against user-defined thresholds.

Calibration tests have shown that Ollaya's error rate is significantly lower (0.081) compared to Jev's (0.246), demonstrating its superior performance.