How do I choose which AI model answers?
Click the model name next to the message box and pick one. Auto is first in the list and picks for you; Claude, GPT and Perplexity models are there too. The choice lasts for the session, and Credits are charged for the model that actually answered.
Pick a model
- In chat, click the model name shown next to the message box.
- Choose from the list — Auto, Claude and GPT models, and Perplexity for sourced search.
- Send your message. The reply comes from that model, and it stays selected for the rest of the session.
Some tasks pick for you: Quick Search and Deep Research switch to a Perplexity model because they need live sources.
Auto
Auto is the first option in the list, marked Recommended. Choose it and Sagelyn picks the model for you, based on the task and the Agent rather than the wording of your message:
- Quick Search and Deep Research use a Perplexity model, because they need live sources.
- The AI Visibility Manager uses a more capable model.
- Everything else uses Claude Sonnet 5.
Every reply is labelled with the model that answered it, for example “Answered by Sonnet 5”, and Credits are charged for that model.
The Agent's default
Each Agent has a default model — set on its Edit page — that every new conversation starts with. Choosing a model in chat changes this conversation, not the Agent's default. Agents you have already set up keep the model they have; Auto does not replace it.
Which model should I use?
Start with Auto, the first option, marked Recommended. It picks the model for you, and each reply shows which one answered. Choose a specific model when you want that one, for example a larger model for strategy or long-form drafts. Quick Search and Deep Research switch to a Perplexity model for you.
Does switching models change what the Agent knows?
No. Instructions, knowledge files and your business memory are the same whichever model answers. The model is the brain; the Agent is the job.
Do different models cost different amounts?
Yes. Credits are charged for the model that produced the reply, so larger models cost more per message.