Hermes Agent API multi-round dialogue is always like "amnesia", many times it's not that the model is stupid, but that you connected the wrong interface. The official API Server documentation clearly states that /v1/chat/completions is a stateless interface, and the full context must be brought by the client itself through messages each time.
The most common misconceptions
Many front-end can reply normally with a sentence the first time, but only send a "continue" sentence in the second round, and the Hermes Agent seems to have forgotten the previous sentence. It's not that it doesn't save, but that you're calling the Chat Completions path, which doesn't inherently protect sessions for you.
How to repair
- If you continue to use
/v1/chat/completions, bring the full message history with you every time. - If you want the server to catch the context for you, change it to
/v1/responses. - Don't misinterpret "OpenAI compatible" as "the server automatically saves all multi-round states".
In a word: multiple rounds of total breaks are not necessarily due to Hermes Agent errors, first check if you are still using stateless Chat Completions.
Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.
