Do you guys have any plans on integrating Jev as a model that we can build with? Might open up some interesting products if we can use it. Thoughts?
AbsoFnlutely should happen - it seems complicated. I see it as an extension to, maybe a post-process for Wingman: “Wingman, review the transcripts for this Agent, and create and insert Jev checks that should be called before calling a full LLM”. Ironically, it may require an LLM call to determine which of the Jev checks might be sufficient at each step. Still, I think that something like DeepSeek Flash 4.1 could do this cheaply and quickly, and the net would be improvements in speed, reproducibility (yes/no), and cost ESPECIALLY if you are using higher cost (Anthropic) models as your main model. Challenge to the team, starting with @nathaniel