# For development (CPU)

> Build and test a client on any machine, with no GPU. The CPU stand-in serves every route of decisio from a small model.

Source: https://opendecisio.com/docs/deploy/development/

Start it from a clone of the release, on any machine.

 

```
git clone --branch v0.18.1 https://github.com/aminry/decisio
cd decisio
uv sync --extra dev --frozen
uv run python -m decisio.serve.vllm_engine --backend hf --model Qwen/Qwen3-0.6B-Base
```

 

Run against the CPU stand-in at v0.18.1 on 2026-10-19: the clone is the pinned release, as the check has it; the install and the server start.

 

## 1 What you get

 

- Every route the real server has, so a client, a [task](https://opendecisio.com/docs/glossary/#task) registration and its tests run against it unchanged.
- Answers from `Qwen/Qwen3-0.6B-Base`, a small model for development; they are not decisio’s measured answers.
- A server that is ready 4 to 8 seconds after it starts, once the model is downloaded, and answers a first [question](https://opendecisio.com/docs/glossary/#question) in about 0.3 seconds on top (on Apple silicon, 2026-10-14).

 

The first start downloads the model, about 1.2 GB, once.

 

## 2 Send it a request

 

It is ready when `/health` answers:

 

```
curl -s http://127.0.0.1:8000/health
```

 

Run against the CPU stand-in at v0.18.1 on 2026-10-19. It printed, trimmed to the fields this page uses:

```
{
 "ok": true,
 "base": "qwen3.6-35b-a3b"
}
```

 

The request from [Get started](https://opendecisio.com/docs/#3-ask-your-first-question) works as it is; only the numbers in the answer differ. [Call it from your code](https://opendecisio.com/docs/use-from/) has the same request from Python, JavaScript and TypeScript.

 

## 3 Teach it a task

 

The walk-through of [Teach a task](https://opendecisio.com/docs/tasks/) runs on the stand-in, end to end: register a task, ask it, list it and remove it. The site checks every one of its commands this way.

 

## What next

 

- [On a cloud GPU](https://opendecisio.com/docs/deploy/cloud-gpu/): serve the default model, Gemma 4 31B, for real answers.
- [On a Mac or a laptop](https://opendecisio.com/docs/deploy/mac/): run a real model on Apple silicon or with Ollama.
- [Call it from your code](https://opendecisio.com/docs/use-from/): the same request from your language.