Ollaya
Ollaya is open-source software that runs decision models on your machine and returns typed, calibrated answers via a local API, for agent developers.
No visits from OpenBeat in the last 30 days yet. Every click on “Visit” is counted here.
Pricing and API checked by an OpenBeat editor on Nov 12, 2026 · facts from the maker’s own site ·
Tool insights
Ollaya runs open decision models on your own hardware and answers typed questions with calibrated probabilities instead of generated text. It exposes a self-hosted, Jev-compatible API, so existing code can point at a local endpoint.
Many agent pipelines call a large language model just to classify something or make a yes/no decision, which adds cost, latency and free-text output to parse. Ollaya moves those decisions to a local model that returns a structured answer in one pass.
AI agent developers and teams that want to replace LLM classification calls with a local, self-hosted service. It is probably not for anyone who needs text generation or who wants a hosted service with no setup.
What it does
Ollaya is an open-source tool by Mert Cobanov. It runs open decision models on local hardware. You send a typed question and get back calibrated probabilities. It does not generate text: each answer comes from a single forward pass through the model.
The main use case is replacing LLM calls that only classify or decide. Examples are routing a request, filtering content or checking a claim. The maker's site lists these models: winnow, laya, decider, kev, nli and gliclass. Per-model descriptions are not included in the material reviewed here, so check the docs.
Key features
- Runs open decision models on your own machine.
- Typed questions in, calibrated probabilities out.
- Single forward pass with no text generation.
- A self-hosted /v1 endpoint that is Jev-compatible, plus native /api endpoints and a CLI.
- Desktop app and command-line installers.
- NVIDIA CUDA and Apple GPU (MLX) support, and CPU-only operation for all models.
- A Modelfile that lets you bake question sets into a model.
- An MCP server and an agent skill.
Pricing
Ollaya is free and open source under the Apache-2.0 license, as of November 2026. The maker's site does not state any paid plans, and no pricing page is listed. You pay only for your own hardware and electricity.
API and integrations
Ollaya offers a REST API. The /v1 endpoint is self-hosted and Jev-compatible, and native /api endpoints are also available. No other SDKs are stated.
Listed integrations are LiteLLM, LangChain, AI SDK, DSPy, LlamaIndex, n8n and Dify. There is also an MCP server for Claude Code, Claude Desktop and Cursor.
Supported platforms are macOS, Windows, Linux, Linux ARM64, WSL 2, Docker and the command line.
Good to know
- NVIDIA GPUs need driver R525 or newer.
- On a Mac, only laya and nli run on the Apple GPU. The other models use the CPU.
- Windows and Linux desktop apps without command-line use run on the CPU.
- Linux ARM64 has no desktop app.
- The latest release on GitHub is v0.18.0, dated 25 October 2026. The repository was last pushed to on 10 November 2026 and is not archived.
- Documentation is at ollaya.dev/docs. A changelog link and a founding date are not stated on the maker's site.
- Because models do not generate text, Ollaya will not replace a chat or writing model. It fits decision steps inside a larger system.
FAQ
Is Ollaya free?
Yes. Ollaya is open source under the Apache-2.0 license and free to use, as of November 2026. The maker's site does not list paid plans.
Does Ollaya need a GPU?
No. All models can run on the CPU. NVIDIA GPUs with driver R525 or newer are supported, as is Apple GPU (MLX), though on a Mac only the laya and nli models use the Apple GPU.
Does Ollaya have an API?
Yes. It provides a self-hosted, Jev-compatible /v1 endpoint, native /api endpoints and a CLI. The SDK listed is REST.
Which tools does Ollaya work with?
The maker lists LiteLLM, LangChain, AI SDK, DSPy, LlamaIndex, n8n and Dify. An MCP server covers Claude Code, Claude Desktop and Cursor.
Which platforms does Ollaya run on?
It runs on macOS, Windows, Linux, Linux ARM64, WSL 2 and Docker, and has a command-line interface. Linux ARM64 has no desktop app.
Releases
No tracked releases yet.
In the news
No coverage yet.