Hermes: model or agent? Choose the right setup

Hermes can refer to a model or to Hermes Agent. A Hermes model generates responses through a serving provider. Hermes Agent is software that runs conversations and tools, keeps context, and connects to a model provider. Decide which one you mean before changing a URL or installing a package.

For the agent-to-router connection, use the dedicated Hermes Agent setup guide. If you are selecting a model named Hermes, keep reading: model selection and agent configuration solve different problems.

Identify what you are trying to run

What you have	What it means	Your next step
A model card under NousResearch	Model weights or documentation	Choose a compatible serving environment
The NousResearch/hermes-agent repository	Agent application	Install the agent and configure its provider
A Hermes model in a hosted catalog	An inference endpoint offered by that service	Check the model ID and your account access
A 9Router model or combo called Hermes	A configured route name	Inspect the route’s actual target

The Nous Research model organization and the Hermes Agent repository are useful starting points. A search result using “Hermes” without an owner, repository, or model card is not enough to choose an install command.

Using a Hermes model through 9Router

9Router is a gateway, not a model-weight runtime. Adding a model’s name to a client configuration does not download its weights or create a GPU server. You need a provider or local serving system that actually exposes the model, plus any required credentials.

First confirm the provider’s endpoint and model identifier. Add that provider using a supported 9Router connection, then inspect the model IDs exposed by your router. Select the identifier from that local catalog in the client. If the requested model is not exposed, fix the provider connection or mapping before changing the client repeatedly.

A hosted route sends prompts to the selected provider. A local gateway does not automatically keep inference local. If local execution matters, inspect the entire path, including fallback and any tools that use remote services.

Running Hermes Agent with a different model

Hermes Agent is not restricted to Hermes models. Its provider documentation describes multiple providers and a custom endpoint option. The agent is the workflow layer; the chosen model is its inference backend.

Use hermes model outside a running chat to configure a provider. The /model command inside a session is for switching among configured choices. If you have not yet added the 9Router endpoint, opening a session’s model menu may not complete that setup.

Start with a model that supports the tool behavior your agent needs. A route that answers a one-line chat might still fail when asked to return a tool call. Long context, image input, and structured output also depend on the selected model and the gateway’s translation path.

What should you check before selecting a model?

Check its model card, license, context limits, serving format, and intended uses. For a hosted model, also read the provider’s current pricing and data-handling information. Those properties belong to a specific model and service; they cannot be inferred from the word Hermes.

Use a small test task with known output: ask for a short explanation, then a harmless file-listing action in a test directory. Review the tool request and result, not just the final response. Keep a record of the model ID and provider so a later update can be compared with the same setup.

Which guide should I read next?

Choose Hermes Agent with 9Router to configure the agent. Choose the proxy guide to check an existing endpoint. If you are deciding whether to run a local gateway at all, compare it with OpenRouter first. Avoid installing an unrelated package simply because its name matches your search.