Overview
When planning a Hermes installation, separate the system into two decisions.
First, choose the brain. Select the primary model that will provide reasoning and generation. A fallback model can be configured so the agent remains usable when the primary option is unavailable or its usage allowance is exhausted.
Second, build the work environment. Decide where Hermes will run, which channels it will use, what applications it may access, what memory provider it will use, how its identity will be defined, how projects will be tracked, and which actions require verification.
This separation also makes troubleshooting easier. A weak response may be a model-selection problem.
A failed calendar action may be a connector or permission problem. Forgotten preferences may be a memory problem. Missing mobile replies may be a gateway problem. Treating all failures as "the AI is bad" hides the actual layer that needs repair.