The Forecasting Company

The Forecasting Company is a research lab that builds systems designed to understand time, starting with a forecasting foundation model. It addresses the challenge of generating accurate predictions from raw historical data without requiring manual feature engineering or fine-tuning. The model reads data such as hourly grid load, prices, or demand for SKUs, and returns a calibrated distribution of what happens next.

The core offering is t0-beta, an open-weights forecasting foundation model accessible via an API. Users describe their data once, and the API handles reshaping. The system can produce up to 200,000 forecasts per second, enabling comprehensive backtesting. A demo tool called Retrocast allows users to test the model on various use cases including energy load, energy price, retail demand, and workforce forecasting. The model is designed to work across domains without per-domain training or per-customer variants.

The product is aimed at organizations needing to forecast time-series data, with users at companies in energy, streaming, retail, and research. The t0-beta model is available with open weights. The company offers an API with documentation, and a Retrocast demo for exploration.