I maintain JarvisCore https://github.com/Prescott-Data/jarviscore-framework, an open-source Python framework for multi-agent systems. When Jev launched I started poking at it, and ended up shipping

I maintain JarvisCore https://github.com/Prescott-Data/jarviscore-framework, an open-source Python framework for multi-agent systems. When Jev launched I started poking at it, and ended up shipping native support in our 1.12 release today . It now drives four things, all opt-in: typed agent decisions, which specialist subagent takes a task, task complexity to model tier, and RAG passage classification after vector retrieval. The RAG one is my favourite. Once FAISS returns candidates, Jev sorts them into accepted / conflicting / excluded, including passages that contradict the question's premise or carry a prompt injection. Excluded text never reaches the generating model. https://jarviscore.developers.prescottdata.io/concepts/decision-models/