Skip to content
Hermes SwitchUI

Switching models mid-chat

Switching models mid-chat

You can change the model used for a conversation at any point without losing your session history. The model selector lives in the chat composer, below the message input area.

[SCREENSHOT: Chat composer with model selector button visible, showing current model name]

What you see

The composer bar includes a model selector button that shows the name of the currently active model for this session. Clicking it opens a dropdown listing available models from the connected provider.

How to switch models

  1. Open a chat session.
  2. Click the model selector button in the composer. The dropdown lists available models.
  3. Click the model you want to use.
  4. The button label updates immediately. Your next message will use the selected model.

The conversation history remains intact. Previous messages were generated by the prior model; new messages use the newly selected model.

Per-session model selection

Model selections are per-session. Each chat session stores its own model choice, persisted in localStorage keyed by the session ID. Switching models in one session does not affect other open sessions.

When you start a new session it inherits the global default model from the agent config (~/.hermes/config.yaml). The per-session override applies only to the current session.

Pinned models

Frequently used models can be pinned for quicker access in the dropdown. Pinned models appear at the top of the list. Pinning is stored locally in the browser.

Auto model selection

Some providers (notably OpenRouter with the auto model ID) support automatic model routing, where the provider selects the best model for each request. Select auto from the model list to enable this. The composer label will show auto.

Limitations

  • Model switching may be blocked in certain gateway modes. If the selector shows a disabled state, the gateway is operating in a mode that restricts model changes.
  • The model list is fetched from the /api/models endpoint, which returns only the curated set of models for the active provider. Not every model the upstream provider offers is listed.