Improved Assistant

Assistant engine setting and Jev accuracy

A server setting can make TypeSafe Jev the Dashboard assistant’s engine instead of OpenAI, and Jev answers count questions and saved-result follow-ups more accurately.

Setting WORKSPACE_ASSISTANT_ENGINE=jev makes the Dashboard assistant use TypeSafe’s Jev for every request instead of OpenAI. The default remains OpenAI. Help describes what Magpie sends to Jev and what Jev cannot do: it chooses operations and values but does not write text, so requests to draft new text are not completed.

Jev now:

  • answers “How many … do I have?” with the product area’s exact count instead of a search;
  • uses saved-result operations for requests about results saved earlier, such as “the To-dos in both saved lists”;
  • does not use one saved result’s identifier as a record identifier in the same step;
  • keeps record names in contact searches, so replies can name the contacts found.

A conversation can move between the two engines.