{
  "id": 9850715,
  "title": "Ollaya – Ollama for open-source, Jev-style decision models",
  "url": "https://urgent.news/2026/09/25/ollaya-ollama-for-open-source-jev-style-decision-models-9850715",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-25T18:33:50.000Z",
  "source": {
    "name": "Hacker News Best",
    "slug": "hacker-news-best",
    "url": "https://ollaya.dev/"
  },
  "original_language": "en",
  "account": "Ollaya is an open-source decision model platform that allows users to ask typed questions about any text or JSON and receive calibrated answers in milliseconds. The platform operates privately and on the user's own hardware, eliminating the need for token-by-token generation. A single forward pass enables Ollaya to answer five questions from Laya, an English model, in approximately 10 milliseconds through the HTTP API.\n\nOllaya serves two primary endpoints: /v1/systemone and /v1/models, adhering to TypeSafe's request and response shapes. The official TypeSafe Python SDK 0.7.1 works seamlessly with the local server. Laya, developed by Convai Innovations, includes an English model, a 100+ language model, a model fine-tuned for typed decisions, and a router to select the appropriate model.\n\nAdditional open decision models, such as von and GGUF LLM-based decision models via llama.cpp, are planned for future releases. Sensitive data, including tickets, emails, and user messages, is scored where it already resides, minimizing exposure. Ollaya runs on the user's machine using ONNX Runtime, either on the CPU or an NVIDIA GPU, with the server listening on 127.0.0.1 by default.\n\nWeights for the decision models are sourced from their authors' Hugging Face repositories, pinned to a specific commit and checked against the SHA256 hash. Ollaya refrains from re-hosting the weights and operates under the Apache-2.0 license, ensuring no metering or API billing. Laya's calibration error (ECE) is 0.081 after temperature fitting, significantly lower than Jev's 0.246, providing more accurate decision-making. A desktop application and command-line interface are available for macOS, Windows, and Linux, while a Docker image is provided for servers.\n\nEvery decision model in Ollaya runs on the CPU; however, on NVIDIA GPUs in Linux, WSL 2, or Docker, response times drop to milliseconds. NVIDIA GPUs require drivers R580 or newer, with CUDA libraries fetched automatically by the installer when detected. For users with AMD and Intel GPUs on Apple, macOS, Windows, Linux, and Docker, models run on the CPU.",
  "summary": "Article URL: https://ollaya.dev/ Comments URL: https://news.ycombinator.com/item?id=49848269 Points: 237 # Comments: 77",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Hacker News",
        "title": "Ollaya – Ollama for open-source, Jev-style decision models",
        "url": "https://urgent.news/2026/09/25/ollaya-ollama-for-open-source-jev-style-decision-models",
        "published": "2026-09-25T18:33:50.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}