Why is Hermes Agent still migrating to the cloud model offline?

The Hermes Agent master model uses a local model, which does not mean that all tasks are offline. The official Fallback Providers documentation is clearly written: vision, web extract, compression, session_search, skills_hub, MCP, and memory flush all have separate auxiliary provider parsing chains. This means that bypass tasks such as compression, memory organization, and web page extraction may still go through OpenRouter or other cloud providers.

How to confirm if it is adjusting clouds

1. See which type of task triggered the request in the log: compression, flush_memories, web_extract, or vision.
2. Check if auxiliary.*.provider is still auto.
3. Check that compression.summary_provider specifies a cloud model.
4. Change the critical secondary task explicitly to main or customize the local endpoint and test again.

If you want to be strictly offline, you can't just change the top-level model.provider, but also close the auxiliary tasks together. Otherwise, the main conversation is localized, and background summaries and memory refreshes may still be cloudy.

In a word: When deploying Hermes Agent offline, you need to check the model, auxiliary, and compression configurations at the same time.

Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.