Hermes Model Guide
Hermes Agent + GPT-5.6: Choose Sol, Terra or Luna
GPT-5.6 gives Hermes Agent three distinct OpenAI options: maximum capability, a balanced default, and efficient volume. This guide explains where each fits and how to connect it without confusing the model with the agent framework.
GPT-5.6 and Hermes Agent are different layers
Hermes Agent is an MIT-licensed agent framework from Nous Research. It supplies the workspace, tools, Skills, memory, browser tooling, scheduling, API surface, and chat connections. GPT-5.6 supplies the language-model intelligence. Keeping those layers separate means you can change models without rebuilding the assistant.
Sol vs Terra vs Luna
| Variant | Hermes Agent role | Recommended use |
|---|---|---|
| Sol | Deep planner | Complex code, research, browser tasks, difficult tool chains |
| Terra | Daily operator | Inbox work, documents, team assistance, mixed tools |
| Luna | Fast worker | Extraction, routing, tagging, repetitive automations |
Connection checklist
- Create or deploy a Hermes Agent instance.
- Choose OpenAI or an OpenAI-compatible gateway as the provider.
- Add your own provider credential through the secure configuration flow.
- Pick the exact GPT-5.6 model ID returned by that provider.
- Run a tool-use test before granting broader permissions.
Hermes Agent's native OpenAI-compatible API is an output surface for applications connecting to your agent. Your upstream OpenAI provider is a separate setting. Do not confuse the agent's API base URL with the upstream model endpoint.
What stays when you switch models?
Your Markdown workspace, SQLite full-text memory, installed tools, Skills, scheduled tasks, and channel connections remain attached to Hermes Agent. Only the model used to reason and produce responses changes. Vector memory is available through optional plugins rather than being part of the native Hermes memory layer.
Cost and permission strategy
Use Sol for tasks where a better plan can prevent expensive mistakes. Use Terra as the default when the assistant handles varied daily work. Use Luna for repeated structured jobs. Regardless of model strength, keep consequential actions reviewable and let AI prepare purchases rather than complete payments with stored credentials.
OpenClaw Launch can host Hermes Agent with server management and channels included. Lite costs $3 for the first month and renews at $6/month; provider usage is separate when you bring your own key.
Hermes Agent and GPT-5.6 FAQ
Does Hermes Agent support GPT-5.6?
Yes. Hermes Agent can use OpenAI and OpenAI-compatible providers, so GPT-5.6 can be selected once the provider exposes the model.
Is GPT-5.6 the same as Hermes models?
No. GPT-5.6 is an OpenAI model family. Hermes Agent is the Nous Research agent framework and can run on models from multiple providers.
Can I switch variants later?
Yes. Your workspace, memories, tools, and channel configuration stay in place when you change the underlying model.