Nixtla — TimeGPT time series forecasting

★ 4K

Nixtla is a Python SDK (the nixtla package) for accessing TimeGPT-2.1, a pretrained foundation model for time series forecasting and anomaly detection. The model is called through the Nixtla API with a key; the repository holds the client code, and no license is declared in its metadata. Reach for it when you need a quick forecast of sales, electricity demand or traffic without training your own model; when you want to catch anomalies in operational metrics or financial series; or when your data lives in Snowflake and you want forecasting there. Inside are NixtlaClient with forecast() and detect_anomalies_online(), cross-validation, exogenous variables, multi-series forecasting, prediction intervals, irregular timestamps, and fine-tuning on your own data. Install with pip install nixtla and get a key on sign-up; for Snowflake there is the nixtla[snowflake] extra and an install script for stored procedures and UDTFs, where by the authors' description data stays inside Snowflake, though an API key is still required. The API is also reachable from other languages including JavaScript and Go. The detect_anomalies() method is deprecated and will be removed in version 1.0. This is a forecasting tool, not financial advice: decisions stay with the user. Not suitable if you need fully offline operation with no external API.