Choose your machine

Decisio serves the same routes from every path; what changes is the model behind them and the memory it needs. Find what you have below, and follow the link.

Figure · Your machine

One server, five ways to run it; start from what you have.

• Linux, one NVIDIA card96 GB, as measuredGPU serverThe default base, Gemma 4 31B.
• Linux, one NVIDIA card, Docker96 GB, as measuredDockerThe same server in a container.
• A Mac with Apple siliconeach base has its own, in Run itMac with MLXThe bases run from their MLX conversions.
• A laptop with Ollama8.6 GB for the Gemma base's smallest tagOllamaA 16 GB laptop fits it; Ollama applies no calibration.
• Any machine, for developmenta small Hugging Face model on the CPUCPU stand-inServes the routes; not the measured system.

The memory is what the README states for each path; the sizes of each base are in Choosing a base.

Source: the README's "Install and run" table.

TL;DR

Whatever you have, the CPU stand-in runs. It serves the routes from a small model on the CPU, so it is for development and not the measured system; for a real base, find your machine in the figure and follow its link.

uv sync --extra dev --frozen
uv run pytest
uv run python -m decisio.serve.vllm_engine --backend hf --model Qwen/Qwen3-0.6B-Base

Run against the CPU stand-in at v0.11.1 on 2026-10-14: the install and the server start; the unit tests passed in decisio's own CI on this commit.

The stand-in serves the routes from a small model on the CPU for development; it is not the measured system.

The five paths

Machine	Path	Memory needed
Linux with one NVIDIA card	GPU server	a 96 GB card, as measured; each base's weights are in Choosing a base
Linux with one NVIDIA card and Docker	Docker	as the GPU server
Mac with Apple silicon	Mac with MLX	each base's, in docs/running.md
Mac, Linux or Windows with Ollama	Ollama	8.6 GB for the Gemma base's smallest tag, so a 16 GB laptop; each tag's size is on its model page (decisio-gemma, decisio)
Any machine, for development	CPU stand-in	a small Hugging Face model on the CPU

docs/running.md has the details of every path.

What next

• Get started: send a first question to the CPU stand-in, on any machine.
• Choosing a base: which of the three models to serve, and when.
• Flags and options: every flag and its default per base.