Okay, this is a big update and shifts how Hermes learns from experience, not just how it stores memory 🤯 Hermes just changed what it does with the things you teach it while it works. Hermes is moving from storing lessons to organizing them by context, so what it learns can come back in the right place, for the right kind of work, at the right time. Task-specific lessons are now meant to improve the Skill for that work instead of filling up global memory. So if you correct Hermes while it is doing something you repeat often, like writing, coding, research, or running a workflow, Hermes is now instructed to put that reusable lesson into the relevant Skill. Think: Reusable lesson for a type of task → Skill Stable fact that should matter everywhere → Memory Temporary detail for this conversation → Session history For example: “Always format these reports with a summary first.” That belongs with the reporting Skill. “I prefer concise responses.” That can belong in persistent memory if it applies everywhere. “Use this temporary staging URL today.” That can stay with the session. The reason this matters is simple. Skills only need to load when they are relevant. Global memory follows Hermes much more broadly and has a limited budget. So instead of every task-specific correction competing for space in memory, the lessons Hermes learns from doing a specific kind of work can make that Skill better the next time it needs it.