Updating model with target values delayed

[eshnEnergiDanmark]

So if I have a MLForecast model where the target data is 7 days, however very explainable to the model, so I use the lags of 168, seasonal rolling means on the 168 lag etc (hourly data). When fitting the model, I look historically, so no problems. However, when predicting in real time, I can update the MLForecast model with data up until 7 days, but then i want to forecast from now and until the range of my future exogenous variables. I have a problem here since the models' last_dates is 7 days back, and then when foreacsting recursively, the error propagates. I just want the model to predict from now and onwards, using the lags of 168 etc. How to achieve this ?