Assistant — Chat model — Mediator

Assistant — Chat model

The AI you're actually chatting with with the Assistant. You can switch at any time — the next turn uses the new model. Doesn't affect your mediator in real conversations.

This is the model that answers you in the AI Assistant. Every Enter runs one call against this model.

Key behaviour

  • Switch any time. The dropdown persists in a server-side session key (assistant_model); the next turn uses the new model. Answers that are already on screen keep the label of the model that produced them (small "· model-name" text under the bubble).
  • Same provider, same key. If the AI Provider section above is set to Anthropic, this list shows Anthropic models. You do NOT need a separate API key — your saved one drives every model.
  • Does NOT affect the mediator. In real customer chats, the mediator uses the model you set in the "Model" field above. This picker is assistant-only.

Suggestions

  • Heavy reasoning / tool-use test: Anthropic Sonnet, OpenAI o-family, Google Gemini Pro.
  • Quick iteration, many small queries: cheaper / faster models — Haiku, gpt-4o-mini, Llama on Groq.
  • If a call fails (rate limit, timeout), the reply comes back empty. Either switch models, or shrink the history before the next turn (🗑 Clear history).

If the dropdown is empty

It loads from /settings/ai/models. If nothing shows up, click the "↻ Fetch models" button in the AI Provider section above — it validates the API key and refreshes the list.