Three different parties are involved, with clearly distinct roles. Nous Research, which publishes the open-source Hermes Agent software, has no access to the content of your exchanges — we run their code on our own infrastructure, and your conversations are never transmitted to them. OpenRouter, the gateway that routes each message to a language model, and the model provider itself — currently OpenAI — do see the content of your exchanges, since that content is what lets them generate a response.
Concretely, OpenAI doesn't use exchanges submitted through its API to train its models — a standard policy of its business offering, different from what applies to consumer ChatGPT usage. Exchanges are, however, retained for a limited time, for security and abuse-detection purposes, before deletion. The model actually active on your account is always visible in your client dashboard (AI Model section).
