Hermes chooses a model and provider, and beginners use hermes setup first, and choose one of Nous Portal and OpenRouter. If you already have ChatGPT OAuth, look at OpenAI Codex; If you have a self-built service that is compatible with OpenAI, choose Custom Endpoint. The good thing about Hermes is that hermes model can be cut and won't lock you into one or the other.
How do you understand the four types of providers?
| Provider | Who is it for? | Key words |
|---|---|---|
| Nous Portal | People who want to run through quickly | zero-config, subscription-based |
| OpenRouter | People who want to choose a model | 200+ models, pay-as-you-go |
| OpenAI Codex | People who are already using ChatGPT | OAuth、Codex models |
| Custom Endpoint | There are people who build their own reasoning services | OpenAI-compatible API |
If your goal is to "use it first", Nous Portal is the most worry-free; If your goal is "multi-model comparison", OpenRouter is more flexible; If you have your own vLLM, SGLang, or other compatible interfaces in your company, Custom Endpoint is more suitable. Hermes also supports provider routing, which allows you to route different requests to different underlying providers, especially in OpenRouter scenarios.
The most misunderstood point
Choosing a provider is not the same as choosing an IQ. Hermes itself doesn't suddenly become smart just because you change providers, the difference is mainly in model capabilities, price, response speed and stability. What you want is not "the most expensive", but "the most suitable for the scene".
Who is it not suitable for? If you only want to run a fixed model offline locally, this article is not the focus, you should first look at the configuration of the local model and inference service.
