Haute

Open-source pricing engine for insurance teams.

Documentation · Getting started · GitHub

---

Haute is a free, open-source pricing engine. Build rating pipelines in a visual editor, train and score models, optimise prices, trace how any price was calculated, and deploy the result as a pricing API, all on standard open-source tools such as Polars, MLflow and Git.

Why Haute

The rest of data science runs on open-source tools, version control, automated testing and continuous deployment. Pricing teams are often still working across a chain of proprietary products, and that stack holds them back:

• Closed and expensive. Licences often run to six figures, the work is stored in proprietary formats, and the calculations can't be inspected, tested or version-controlled like code. Leaving means rebuilding everything.
• Split across tools. Data preparation, modelling, rating structures, optimisation and the live rating engine often sit in separate products. Every change is exported, re-keyed and reconciled between them, and each hand-off is a chance for error.
• Manual releases. Changes reach production through exports, spreadsheet checks and email sign-offs rather than automated tests, staged deployments and impact analysis.
• Skills locked to one vendor. Analysts learn an interface instead of Python. When the tool can't do something, neither can the team, and newer modelling methods arrive only when the vendor adds them.

Haute replaces that chain with one open pipeline:

• Open, and yours. Haute is free and open source. Each pipeline is an ordinary Python file with its settings in JSON files beside it, kept in sync with the editor as you work. You can read, test and version pipelines like any other code, see exactly how every price is calculated, and run them on your own infrastructure.
• One pipeline from data to deployed price. Data preparation, rating, models and optimisation live in one pipeline that analysts build on the canvas and developers edit as code. The same logic prices batch data and live quotes, so nothing is rebuilt for production.
• Releases with guardrails. One wrong factor can misprice a whole book, so every change is validated, deployed to staging and approved in CI/CD before it reaches production. Analysts push their changes; they don't need Docker or cloud tools on their machines.
• Skills that transfer. Everything underneath is standard tooling, and the step builder shows the Polars code it writes, so what analysts learn applies well beyond Haute.

Haute is for pricing and actuarial teams who want to build, test and release their own pricing changes without waiting on engineering, and for teams moving off legacy platforms who want to keep their guardrails.

Built on open-source tools

Haute connects standard tools rather than replacing them:

Tool	Role in Haute
Python	Every pipeline is a Python file, with one function per node.
Polars	The data engine: columnar, multi-threaded and strict about types.
MLflow	Experiment tracking and the model registry, locally, on your own server or on Databricks. Trained models are logged with their model cards, and pipelines score registered models by version or alias.
Git	Version control. Every save in the editor is a commit, so a pipeline's history can be reviewed like any other codebase.
CatBoost, XGBoost, LightGBM, RustyStats, InterpretML	Model training (gradient boosting, GLMs and explainable boosting machines), with Optuna for hyperparameter tuning.
price-contour	Constrained price optimisation.
GitHub Actions, GitLab CI, Azure DevOps	Validation and releases, through workflows haute init generates.
Databricks	Optional: read Unity Catalog tables, track models in Databricks MLflow, and serve pipelines on Model Serving.
Docker	Serve a pipeline from a container image you run on your own infrastructure.

Quick start

uv init my-pricing-project
cd my-pricing-project
uv add haute
uv run haute init
uv run haute serve

uv add needs a project to add Haute to, which uv init creates. haute init scaffolds a project with its configuration, test quotes and CI/CD workflows, and haute serve opens the editor in your browser. The getting started guide covers setup in more detail.

Features

Build pipelines

• Rating steps. Factor tables of one to three factors, banded or raw, with spreadsheet copy and paste, default values, and combined outputs (multiply, add, min or max).
• Banding. Group continuous values with breakpoints, or map categories into rating groups.
• Polars step builder. Filter, derive, join, group, pivot and more without writing code. The steps are saved as Polars in the pipeline, and you can drop into code when you need to.
• Data. Parquet, CSV, JSON and Arrow files, plus Excel, Delta Lake, Iceberg and Databricks Unity Catalog through optional packages. Nested JSON or XML quotes are split into flat tables, and responses can be nested JSON.
• Reuse. Group nodes into submodels, and create instances that apply a node's logic to different inputs.
• Live and batch. Source switches send development or batch data through the same logic that prices live quotes.

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗