Hermes Agent's /model is a model switching command in a session, which is suitable for scenarios where "organize with cheap models first, and then cut strong model decisions". Normally, the cutting model does not empty the current conversation context; It changes the provider/model used for subsequent replies, not to restart the session.
The most common usage:
Enter /model in the Hermes Agent CLI to view the current model and options; If you already know the goal, you can also write /model claude-sonnet-4, /model zai:glm-5, /model custom:qwen-2.5 directly. After v0.8.0, gateway portals such as Telegram, Discord, and Slack also support more natural model switching, and some platforms can also be selected with buttons.
When to cut the model
| Scene | suggested |
|---|---|
| Data collation, summary, compression | Use cheap or free auxiliary models |
| Code modification, complex reasoning | Cut to a stronger master model |
| The current provider is slow or error | Switch provider or other models under the same provider |
| Privacy or cost sensitive | Switch to a self-built endpoint or a model with a clear price |
How to troubleshoot problems after cutting
If the switch fails, check whether the target provider is logged in or has an API key, and then check whether the model name is supported by the current provider. Aggregation portals such as OpenRouter and Nous Portal are usually more flexible; If you build your own custom endpoint, you need to check whether the service supports the corresponding model name.
Don't think of /model as a "IQ button." It changes capacity and cost, but it doesn't automatically fix bad prompts, wrong file permissions, or broken MCP tools. Save the mission-critical state before cutting the model, and let Hermes Agent restate the current goal in one sentence after cutting, which is the most stable.
Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.
