TimeFM — Zero-Shot Time Series Forecasting

Upload or paste any time series. Get a forecast with calibrated uncertainty.

No training, no configuration needed. The model handles seasonal patterns, trends, intermittent demand, and more. Built-in guardrails automatically detect when the model is uncertain and fall back to robust baselines.

Input: A column of numbers (minimum 20 values, ideally 200+)
Output: Forecast with 50%/80%/90%/95% prediction intervals

How to use:

1. Paste comma or newline-separated numbers in the text box, or
2. Upload a CSV file with a numeric column
3. Set the forecast horizon (how many steps ahead)
4. Click Submit

What it's good at:

• Retail demand forecasting (MASE 0.83 on Walmart M5 data)
• Seasonal patterns (MASE 0.69)
• Intermittent demand (MASE 0.56)
• Trend forecasting (MASE 0.77)

Limitations:

• Univariate only (one variable at a time)
• Works best with 200+ data points
• Cannot predict random walks (no model can)