The free, open-source engine for insurance teams. Modern analytical workflows and engineering practices just 30 seconds away.
The whole pipeline runs as one Polars query. Haute works out which columns each step uses, so a wide file is read narrow.
Execution streams through the data a slice at a time and writes each slice out as it goes, so only one slice is ever held in memory.
Execution lets Polars spread each query across all your cores. Large quote files are read in parallel, and RustyStats and Price Contour split GLM fitting and optimisation across cores too.
Edit a step and Haute keeps the results upstream of it. Joins, group-bys and scored batches are saved, so the next preview starts from them.
Train CatBoost, XGBoost, LightGBM, EBM or a GLM from a node in the pipeline, validate with a holdout, cross-validation or a time-based split, and watch the loss curve live.
Optuna searches the boosted models’ parameters across your validation splits, and the best trial’s settings are one click from the final fit.
Price Contour finds the prices that maximise your objective within your constraints, quote by quote or as a ratebook, and traces the efficient frontier so you choose the trade-off.
Each model comes with a validation pack: lift, Gini and Lorenz curve, held-out test metrics, residuals, feature importance, partial dependence and actual against expected by factor; boosted models add SHAP, GLMs their relativities and p-values.
Central repository for your models and their parameters, metrics, diagnostics and model cards. Score any logged run or registered model back in the pipeline.
uv add hautehaute inithaute serve