Google’s BigQuery DataFrames introduced a pandas accessor that invokes cloud AI functions, including predictions without local model training.

The new `.bigquery` accessor connects pandas workflows to BigQuery AI functions. Google’s announcement identifies forecasting, classification, and high-dimensional embedding generation among the supported workflow categories. It describes execution using cloud infrastructure while retaining the familiar pandas DataFrame interface.

The regression example replaces a locally trained modeling pipeline. Google contrasts extracting features, scaling them, and fitting a scikit-learn linear-regression model with a single `ai.predict()` call using historical and new-data DataFrames. The example identifies TabFM as the prediction mechanism and supplies the target through `label_col='target_metric'`.

The illustrated setup still requires data and cloud-project configuration. Importing `bigframes.pandas` enables the accessor, and the example sets a Google Cloud project before reading historical and new data from CSV files. “Without local training” therefore describes avoiding the local model-fitting process, not eliminating historical examples, target-column identification, or cloud configuration.