Models

Pick a model with the model option; leave it off for the default. Every bundled model ships in the zero-dependency core, no ONNX runtime required.

Forecasting

model	Kind	Notes
auto	meta	The default when no model is named. Rolling-origin backtests each candidate per series and forecasts with the lowest-error one. See Auto-selection.
theta-classic	statistical	The Theta method (default-ts aliases to it). Strong, cheap trend + seasonality baseline.
stub-seasonal-naive	statistical	Seasonal-naive with drift. The benchmark floor every forecasting paper reports against.
tsb	statistical	Teunter-Syntetos-Babai, for intermittent demand (many zeros). Forecast-only.

Anomaly detection

model	Kind	Notes
theta-classic	statistical	The default detector: flags points that fall outside the one-step-ahead forecast interval.
stub-seasonal-naive	statistical	The same residual detector over the seasonal-naive forecaster.
sub-pca	statistical	Subsequence PCA (Jacobi eigensolver in C); competitive with SOTA on the TSB-AD-U benchmark. Pick it with '{"model":"sub-pca"}'; its rows carry no interval fields.

Tabular prediction

model	Kind	Notes
native students	native	Trained by fit (gbt or tree) or distill_predict (tree, gbt, or mlp). fit returns a model blob by default, or registers under your id when you pass '{"register":"..."}'; either way the predict scalar serves it per row.
knn5-incontext	statistical	In-context k-nearest-neighbors (k=5), a zero-setup baseline with no training step. Serve it through predict_batch(train_sql, apply_sql).

Distilled forecast students

distill_forecast registers a DLinear/TiDE-style native student under the student_id you choose. Serve it through forecast() with '{"model":"<id>"}', or enter it as a candidate in auto. These are portable blobs: copy the model row to another database and it works there. See Distillation.

Foundation-model teachers (optional ONNX build)

The default build is pure C. make loadable-onnx adds a runtime='onnx' path so distill_forecast/distill_predict can distill from a live foundation-model teacher (Chronos for forecasting, a tabular FM for the tabular path). Serving the distilled student never needs that build. License-tagged teachers enforce PREDICT_ERR_LICENSE unless you pass accept_license.