AI models
Every agent runs on three AI models. Think of them as its brain, its ears and its note-taker. To change them, select Model at the top of the builder. The AI models dialog opens.
| Choice | What it's for |
|---|---|
| Language model | The brain: it decides what the agent says and which arrow to follow |
| Transcriber | The ears: it turns what your contact says into text |
| Extraction model | The note-taker: it picks out details like a name or a date from the conversation |
The voice is chosen separately. See Voices.
Language model
New agents start on gpt-4.1-mini, a good all-round choice. The list begins with it, and the models after it are the other choices.
If your agent is on a model we no longer offer, it keeps working and the list marks that model as unavailable. Choose another one whenever you like.
Transcriber
The transcriber decides how well your agent hears. Pick one that supports the languages your contacts speak. For Indian languages, Sarvam is a good place to start.
Extraction model
This only matters for steps that collect details. Each of those steps can use its own model, or Default to follow the agent's choice. See Steps.
Common questions
Which AI models does an agent use?
Three: a language model that decides what to say, a transcriber that turns speech into text, and an extraction model that picks details like a name or date out of the conversation.
Where do I change the model?
Select Model at the top of the builder to open the AI models window.
Which language model do new agents start on?
New agents start on gpt-4.1-mini. You can pick another from the list at any time.
What happens to an agent on a model that is no longer offered?
It keeps working, and the list marks that model as unavailable. Choose another model whenever you like.