TimesFM Forecasting

TimesFM is a zero-shot time-series foundation model for univariate forecasting. Use it for sales, demand, sensors, vitals, prices, weather, and other temporal numeric series when you need point forecasts plus calibrated quantile intervals without training a custom model.

When to Use This Skill

Use when you need:

• zero-shot forecasting for one or many univariate series
• probabilistic forecasts with prediction intervals
• batch forecasting across many related series
• covariate forecasting with exogenous variables via TimesFM 2.5 + timesfm[xreg]
• anomaly screening by comparing actuals against forecast quantile bands

Do not use when the task needs coefficient interpretation, time-series classification/clustering, multivariate VAR/Granger causality, generic tabular ML, or unavailable optional XReg dependencies.

Mandatory Safety Preflight

Always run the system checker before first model load: