Overview
Welcome to the TFC API. Here we expose our own, as well as open-source models, behind a unified API. The foundation models we currently support are listed below. Please note that in their present implementations, TimesFM and Chronos neither support exogenous variables.
| Model | Description | Covariates? | Forecast? | Retrocast? |
|---|---|---|---|---|
| t_0 | The Forecasting Company's model (coming soon) | ✅ | ✅ | ✅ |
| tfc-global | Global model for time series | ✅ | ✅ | ✅ |
| tabpfn-ts | Foundation model from Uni. of Freiburg | ✅ | ✅ | ✅ |
| moirai | Salesforce's Moirai | ✅ | ✅ | ✅ |
| moirai-moe | Salesforce's Moirai MoE | ✅ | ✅ | ✅ |
| timesfm | Google's TimesFM | Future covariates only | ✅ | ✅ |
| timesfm-2 | Google's TimesFM 2.0 | Future covariates only | ✅ | ✅ |
| chronos | Amazon's Chronos | ❌ | ✅ | ✅ |
| chronos-bolt | Amazon's Chronos-Bolt | ❌ | ✅ | ✅ |
| tirex | NXAI's model based on xLSTMs | ❌ | ✅ | ✅ |
| aarima | Auto-ARIMA | Soon | ❌ | ✅ |
| aets | Auto-ETS | Soon | ❌ | ✅ |
We also support auto-ETS and auto-ARIMA behind our long-running jobs endpoint.
Running forecasts
Our main endpoint for forecasting is /forecast. This returns forecasts directly for each request.
For long-running jobs (e.g. when you may want to run many historic forecasts on the same series at once (we call this retrocasting) for the purpose of backtesting),
we have a separate endpoint /retrocast-jobs where you can start jobs and poll for results given the returned job ID.
Support and feedback
Please reach out at [email protected]