NEWS

nixtlar 1.0.0 (2026-06-23)

This release consolidates all changes since the last CRAN release (0.6.2), including the development versions 0.6.3 and 0.6.4.

nixtlar now requires R (>= 4.1.0).
Set default value for id_col to unique_id in nixtla_client_historic to match the other core functions.

Added support for distinguishing between historic and future exogenous variables in the forecasting and cross-validation functions. This feature was available in the development versions 0.6.3 and 0.6.4, but not in the CRAN version (0.6.2).

Fixed an HTTP 422 error when using future exogenous variables with h = 1.
Updated internal functions to use the GET method, as the POST method is no longer supported.
Fixed in-sample timestamp reconstruction in nixtla_client_historic for series that share some but not all in-sample sizes.
Fixed handling of mixed date-only and datetime character ds values that previously produced "failed to parse" warnings and NA timestamps.
Made output date-format detection robust across the whole ds column in nixtla_client_historic, nixtla_client_detect_anomalies, nixtla_client_forecast, and nixtla_client_cross_validation, so a leading midnight value no longer truncates the time component of the remaining rows.

Development version. Release notes available here.
Added support for distinguishing between historic and future exogenous variables in forecasting and cross-validation functions. Users can now specify which exogenous variables have only historical values and which include both historical and future values.

Updated to support TimeGPT API v2.
Added support for tibbles, in addition to data frames and tsibbles.
Standardized date format requirements to YYYY-MM-DD or YYYY-MM-DD hh:mm:ss, as character strings or date-time objects.
Default id_col set to unique_id, aligning with the Python SDK. If the dataset contains a single series, id_col=NULL should be used. Only character or integer values are accepted for id_col.
Release notes available here.