DOCUMENTATION / Getting started

Documentation

Get Noullama running, work with decision models, and connect your applications to a local API.

Start with a decision.

Noullama is a local workbench and API for native decision models using llama.cpp. Give a model a state and a set of typed questions, then use its answers in your application.

Start in the desktop playground, tune your questions with saved presets, and connect your code through the local API.

YOUR FIRST RUNInstall. Load a model. Run a decision.Go from a Linux package to your first result.

Three ways to ask

choice: Choose one option from a set, such as the team that should handle a support ticket.
noul: Answer a yes/no question, such as whether a customer wants a refund.
score: Rate a state against an ordered set of levels.

Your state is the input to assess: text, JSON, or images with a compatible model and projector. Your questions define what you want to decide.

Find your next step

Set up your runtime

Find a llama.cpp build and choose CPU or GPU inference.

Find a model

Download a GGUF or load one already on your machine.

Work in the playground

Edit questions, run decisions and save your own presets.

Ask about images

Load a projector and add images to your state.

Connect your application

Use native System One requests or the OpenAI adapter.

Keep Noullama up to date

Install updates and opt in to pre-release versions.

Build from source

Set up the toolchain and build a standalone executable.

Find a fix

Check load errors, runtime logs and common setup issues.

Before you begin

Noullama is in early beta. Packages are available for 64-bit x86 Linux as .deb and .rpm. There are no macOS or Windows builds yet.

You will need the Hugging Face CLI for downloads and a compatible llama.cpp runtime. Noullama can download the runtime for you; it does not bundle one.