Skip to content

Installing Haute

Make sure you've set up your environment first (VS Code and uv).

Create a project and install Haute

Open the VS Code terminal and run these commands one at a time:

uv init my-pricing-project
cd my-pricing-project
uv add haute
uv run haute init --target databricks

uv run runs Haute from the project's environment, which you have not activated yet. --target names where you will deploy: databricks (the default), container, azure-container-apps, aws-ecs or gcp-run.

This fills the project folder with everything Haute needs: the starter pipeline rating/main.py (the pipeline the editor opens), a rating/utility/ folder for your own helper functions, a data/ folder for your data files, a haute.toml configuration file, test quote templates, CI/CD workflow files, and a .env.example credential template. It also removes the main.py that uv init created, because Haute uses rating/main.py.

Copy the credential template to .env and fill in your values. .env is gitignored, so your credentials are never committed:

cp .env.example .env

Activate the environment and run

uv add already created the project's virtual environment, .venv, and installed Haute into it. Activate it and start the editor:

.venv\Scripts\activate
haute serve

You'll see (.venv) appear in your terminal prompt after the activate step - that means the virtual environment is active. haute serve opens the Haute visual editor in your browser. If you see it, you're all set.

What is a virtual environment?

A virtual environment is a private space for your project's packages. Without one, installing packages would change your whole computer's Python setup. With a virtual environment, each project gets its own isolated set of packages. You activate it each time you open a new terminal.

Installing extras

Depending on your deploy target, you may need additional packages:

uv add "haute[databricks]"         # Adds SQL support and pins Databricks clients

XGBoost GPU training

Haute installs XGBoost's CPU-only build (xgboost-cpu). To train XGBoost models on an NVIDIA GPU on Windows or Linux, swap in the full CUDA build once, then restart haute serve:

haute gpu-setup            # installs xgboost (CUDA) at the same version
haute gpu-setup --check    # reports the build, the GPU and whether GPU training works
haute gpu-setup --cpu      # switches back to xgboost-cpu

The command needs the NVIDIA driver (nvidia-smi must list the GPU) and checks the result in a fresh Python process. The CUDA build is a larger download (about 140 MB on Windows). Re-syncing the project (uv sync) restores xgboost-cpu; run haute gpu-setup again afterwards. macOS has no CUDA build.

Troubleshooting

Managed Windows computers

Some Windows Application Control policies allow an organisation-approved Python interpreter but block generated console launchers such as .venv\Scripts\haute.exe. Haute exposes the same CLI directly through Python, so the console launcher is optional.

If the project already has a blocked .venv, recreate that dependency environment from the approved Python path supplied by your IT team. --clear replaces only .venv; uv sync restores the project's declared packages. The two policy flags prevent uv from substituting a managed download:

uv venv --clear --python "C:\Path\To\Approved\python.exe" --no-managed-python --no-python-downloads
uv sync --no-managed-python --no-python-downloads
.\.venv\Scripts\python.exe -m haute serve

Calling the environment's Python explicitly means activation is optional. Once it is activated, the shorter python -m haute serve is equivalent. Both module forms and haute serve invoke the same command implementation and accept the same options. The module form works for haute init too: python -m haute init. If the approved Python interpreter itself is blocked, IT must permit or provision that runtime; Haute does not bypass operating-system policy.

macOS: XGBoost or LightGBM will not load

The macOS wheels of XGBoost and LightGBM use the system's OpenMP runtime, which macOS does not ship. If training or scoring an XGBoost or LightGBM model fails with an error mentioning libomp.dylib, install it with Homebrew and restart haute serve:

brew install libomp

Windows and Linux installs need no extra step.

haute serve doesn't open anything in my browser

Look at the terminal output for a line like Running on http://localhost:8000. Copy that address and paste it into your browser. If you see an error, make sure your virtual environment is active ((.venv) in your prompt).

(.venv) isn't showing in my terminal prompt

Run .venv\Scripts\activate. You need to do this every time you open a new terminal window.

What's next?

You've got Haute running locally. From here:

  • Build a pipeline - see the Building Models guide to create your pricing pipeline
  • Deploy it - when you're ready to go live, head to the Deployment docs. If you're new to Git, CI/CD, and other deployment concepts, read Before You Start first - it explains everything in plain English.