# Ollaya 发布，本地运行开源决策模型并兼容 TypeSafe API

- 来源：Hacker News 热门（buzzing.cc 中文翻译）
- 作者：Ardakilic
- 发布时间：2026-09-26 04:02
- AIHOT 分数：54
- AIHOT 链接：https://aihot.news/items/cmuhf1x2403gvrojn41jenj54
- 原文链接：https://ollaya.dev/

## AI 摘要

Ollaya 发布，一个开源的本地决策模型运行工具，对文本或 JSON 的类型化问题返回毫秒级校准答案。支持 laya、decider、nli、gliclass 等开放权重模型，兼容 TypeSafe 的 /v1/systemone 和 /v1/models 接口，官方 TypeSafe Python SDK 0.7.1 可直接使用。

## 正文

Ask typed questions about any text or JSON and get calibrated answers in milliseconds. Private, open source, on your own hardware.

DownloadBrowse models

媒体内容 · 前往原文查看

ollaya run laya --preset triage \

"I was charged twice this month and want a refund."

Answers returned by the model

QuestionAnswerProbability

intentrefund1.00

is_urgentno0.87

frustration1.59 / 3 clearly annoyed0.36

refund_requestedyes0.88

churn_riskno0.89

Real output: routed to laya:en, answered in 8.9 ms on an RTX 4090.

Fast

Decisions in milliseconds.

A decision model answers in a single forward pass, with no token-by-token generation. On your own GPU, a five-question request to Laya takes about 10 ms, end to end through the HTTP API.

8–10ms

Laya on Ollaya

RTX 4090, five questions, end to end

236–276ms

TypeSafe Jev

Hosted API, median request

媒体内容 · 前往原文查看

Every model, one scale · median latency, lower is better

laya:multilingual8.1 ms

laya:en9.6 ms

gliclass14.7 ms

nli20.4 ms

decider:0.8b155 ms

decider:2b190 ms

TypeSafe Jevhosted API236–276 ms

Ollaya: median of a five-question request through the HTTP API on an NVIDIA RTX 4090 (laya in fp16, the others in fp32). Jev: median request latency of the hosted API in third-party benchmarks (AbdelStark/jev-benchmarks, nibzard/decision-model-benchmark), which includes the network. Setups differ, so read it as an order-of-magnitude comparison.

Drop-in compatible

Speaks TypeSafe's API.

Ollaya serves /v1/systemone and /v1/models with TypeSafe's request and response shapes. The official TypeSafe Python SDK 0.7.1 works unchanged against a local server.

Request

export TYPESAFE_BASE_URL=http://localhost:11435 export TYPESAFE_API_KEY=local export TYPESAFE_DEFAULT_MODEL=laya

curl http://localhost:11435/v1/systemone -d '{ "model": "laya", "state": "Can I get an invoice for last month?", "questions": { "intent": { "type": "choice", "instructions": "What does the customer want?", "criteria": { "invoice": "Needs an invoice or receipt", "refund": "Wants money back", "other": "Anything else" } } } }'

Response

{ "model": "laya:en", "answers": { "intent": { "type": "choice", "choice": "invoice", "confidence": 0.9547, "probabilities": { "invoice": 0.9698, "refund": 0.0172, "other": 0.013 } } }, "usage": { "input_tokens": 43, "output_tokens": 0 } }

TypeSafe compatibility guide

Open models

Open weights, ready to pull.

Start with Laya from Convai Innovations: an English model, a 100+ language model, a model fine-tuned for typed decisions, and a router that picks for you.

layaOpen decision models from Convai Innovations. Typed, calibrated answers to choice, score and yes/no questions in a single forward pass, in English and 100+ languages.322m · 421m

deciderDecoder decision models by Mapika on Qwen3.5: the answer is read from option-letter logits in one forward pass. The most accurate open decision model Ollaya ships.0.75b · 1.9b

nliZero-shot classifiers by Moritz Laurer: every option becomes a hypothesis scored for entailment. The most accurate encoder model on typed decisions in our tests.396m · 435m

gliclassInstruction-following zero-shot classifier by Knowledgator: all options of a question are scored in one pass, so cost barely grows with the number of options.439m

Browse all models

More open decision models are planned: von, GGUF LLM-based decision models via llama.cpp.

Your data stays yours

Private by default.

Tickets, emails and user messages are often the most sensitive data you have. With Ollaya they are scored where they already live.

Local

Runs on your machine with ONNX Runtime, on the CPU or an NVIDIA GPU. The server listens on 127.0.0.1 by default.

Open weights

Weights come from their authors’ Hugging Face repositories, pinned to a commit and checked against sha256. Ollaya never re-hosts them, and the runtime is Apache-2.0.

No per-token fees

Run as many decisions as your hardware can handle. No metering and no API bill.

Calibrated

Probabilities you can put thresholds on. Laya’s calibration error (ECE) is 0.081 after temperature fitting, vs 0.246 for Jev.

Platforms

Runs where you work.

A desktop app and a command line for macOS, Windows and Linux, and a Docker image for servers. Every model runs on the CPU; an NVIDIA GPU on Linux, in WSL 2 or in Docker takes a request down to milliseconds.

PlatformDesktop appCommand lineGPU

macOSApple siliconDesktop appMenu bar app.dmgCommand lineInstall scriptGPUCPU only

Windows10 and 11, x64Desktop appDesktop app.exe or .msiCommand linePowerShell scriptGPUCPU onlyNVIDIA via WSL 2

Linuxx86-64Desktop appDesktop appAppImage, .deb, .rpmCommand lineInstall scriptsystemd serviceGPUNVIDIA, CUDA 13

LinuxARM64Desktop appNot availableCommand lineInstall scriptsystemd serviceGPUCPU only

WSL 2Linux on WindowsDesktop appNot availableCommand lineInstall scriptSame as LinuxGPUNVIDIA, CUDA 13

Dockeramd64 and arm64Desktop appNot availableCommand lineImage on GHCRGPUNVIDIA, CUDA 13:cuda image, amd64

Install for your platform

NVIDIA GPUs need driver R580 or newer; the installers fetch the CUDA libraries only when they find one. On Apple, AMD and Intel GPUs, models run on the CPU.

Get up and running in minutes.

One binary, one command: ollaya run laya.

Download

macOS, Windows, Linux and Docker · Apache-2.0 · GitHub