Quickstart

1. Install Noullama

Choose the package for your distribution. These commands install v0.2.0 for Linux x86_64 / amd64, downloading directly from GitHub Releases. For newer releases, visit the download page.

Debian, Ubuntu & Linux Mint

curl -fL --output noullama.deb \
  'https://github.com/parasite6/noullama/releases/download/v0.2.0/noullama_0.2.0_amd64.deb' && \
  sudo apt install ./noullama.deb

Fedora & compatible RPM distributions

curl -fL --output noullama.rpm \
  'https://github.com/parasite6/noullama/releases/download/v0.2.0/noullama-0.2.0-1.x86_64.rpm' && \
  sudo dnf install ./noullama.rpm

The package manager installs system library dependencies and asks for your password. The install command only runs if the download succeeds. Open Noullama from your application launcher after installation.

2. Prepare model downloads

Install the Hugging Face CLI using its official setup instructions. Check that it is available:

hf --help

Public models do not require a login. Gated models need access granted on Hugging Face and authentication through an existing login or a session token in Noullama.

You also need a compatible llama-server. Noullama can download it in the next steps. Package users do not need Node.js, Rust or Blaze; those are for building from source.

3. Load your first model

1. Open Hub. Search for a decision model or enter LiquidAI/d1-2B-GGUF, select the repository, and download a GGUF weight file. Already have one? Open Models → Browse local model.
2. Open Runtime → llama.cpp. Select an installed build, download an official build, or enter a custom llama-server path and click Check.
3. Under Runtime → Model, set GPU layers. Use 0 for CPU inference. Leave Vision off for this text-only first run.
4. Click Load model in the model bar. Wait until the model is ready.

4. Run a decision

Open Playground, select a preset under Run settings, and edit the questions and state. Click Run decision. Use Playground & presets to learn how to save your setup.

5. Try the local API

Keep Noullama open with the model loaded. Check the address in Local API; this example assumes the default port, 11435.

curl http://127.0.0.1:11435/v1/systemone \
  -H 'Content-Type: application/json' \
  -d '{
    "state": "I was charged twice. Please refund the duplicate payment.",
    "questions": {
      "team": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {"billing": "Charges and refunds", "technical": "App errors"}
      },
      "refund": {"type": "noul", "instructions": "Is the customer asking for a refund?"}
    }
  }'

For Python and other integrations, continue to the Local API guide.