How Hermes Sense works
Hermes Sense is the intelligence layer inside Hermes. It is designed to stay useful, bounded and explicit.
1. Local-first routing
Hermes Sense prefers local execution first. Connected assist exists as an explicit layer for broader reasoning or fresher answers when you choose it.
2. Context boundaries
Hermes Sense keeps its own thread and memory. It does not silently inspect the rest of your chats. If you want it to use message context, you bring in a snapshot on purpose.
3. Files and voice
Hermes Sense can work with voice, summaries, drafting and file analysis. Supported formats are handled locally first where possible, with narrower connected fallback when needed.
4. Memory and feedback
Hermes Sense can use local memory, user feedback and preference signals to behave more coherently over time. That is product memory and feedback, not a hidden claim of silent model training.
5. Runtime honesty
Hermes distinguishes between local capability, connected assist and actual server-side runtime. It does not claim a stronger connected model is active if the live runtime is different.
6. What Hermes Sense is not
It is not a generic omniscient chatbot, and it is not positioned as an excuse to weaken the rest of the messaging product.