README.md

---
title: README
emoji: 🦉
colorFrom: gray
colorTo: gray
sdk: static
pinned: false
---

# Ollaya

**Run open decision models locally, the way Ollama runs LLMs.** Typed questions in, calibrated answers out, in milliseconds, behind a TypeSafe-compatible API.

```sh
curl -fsSL https://ollaya.dev/install.sh | sh
ollaya run laya --preset triage "I was charged twice this month and want a refund."
```

| Model | Based on | |
|---|---|---|
| [laya](https://huggingface.co/ollaya-dev/laya) | [convaiinnovations/laya](https://huggingface.co/convaiinnovations/laya) | The fastest: 8–10 ms for five questions on an RTX 4090. English and 100+ languages. |
| [decider](https://huggingface.co/ollaya-dev/decider) | [Mapika/decider-2b](https://huggingface.co/Mapika/decider-2b), [0.8b](https://huggingface.co/Mapika/decider-0.8b) | The most accurate. Qwen3.5 decoders. |
| [nli](https://huggingface.co/ollaya-dev/nli) | [MoritzLaurer/deberta-v3-large-zeroshot-v2.0](https://huggingface.co/MoritzLaurer/deberta-v3-large-zeroshot-v2.0), [ModernBERT-large](https://huggingface.co/MoritzLaurer/ModernBERT-large-zeroshot-v2.0) | Zero-shot NLI, the most accurate encoder. |
| [gliclass](https://huggingface.co/ollaya-dev/gliclass) | [knowledgator/gliclass-instruct-large-v1.0](https://huggingface.co/knowledgator/gliclass-instruct-large-v1.0) | Instruction-following zero-shot classifier. |

These repositories hold only what Ollaya derives: weightless ONNX graphs and decision configs. The graphs read the weights from the authors' own `model.safetensors`, pinned to a commit and verified by sha256. Ollaya never re-hosts weights.

[Website](https://ollaya.dev) · [Models](https://ollaya.dev/search) · [Docs](https://ollaya.dev/docs) · [GitHub](https://github.com/ollaya-dev/ollaya)

Apache-2.0. Independent project, not affiliated with Ollama or TypeSafe.