Forecasting API deployment
Integrate an external provider or your own deployment with explicit model, payload, and operating contracts.
The previous TSFM API
TSFM no longer offers a hosted forecasting API. Its former API-key examples and unified inference endpoints are not current integration instructions. The browser playground is local execution and does not replace a server endpoint for scheduled jobs or remote clients.
Choose an external execution service
Verify that a provider supports the selected time-series checkpoint and task, not just a related model name. Check the model revision, credentials, limits, batching, privacy policy, prices, and failure behavior. A custom Hugging Face Inference Endpoint may need a handler implementing model-specific preprocessing and forecast decoding.
Define the boundary contract
Specify history values, cadence, cutoff, horizon, covariate shapes, and requested output types. Return model provenance, issue time, quantile labels, and numerical values in a predictable schema. Test realistic payload sizes, missing values, timeout behavior, and invalid output shapes before treating the endpoint as production-ready.