Skip to main content
Chat Model nodes determine which large-language model the agent uses for its reasoning, responses, and task execution.
Each model node represents a different provider or model family, but they all serve the same purpose: powering the agent’s thought process during a session. These nodes allow you to customize:
  • Model selection — choose the specific LLM used for generation.
  • Advanced settings — such as reasoning depth or provider-specific optimization.
  • Performance tradeoffs — balancing accuracy, latency, and cost depending on your use case.
Chat models can be configured in two places:
  • Start Node — sets the default model for the entire call.
  • Agent Nodes — optionally override the Start Node’s model for specific tool paths.
Below are the supported model providers and their configuration details.

​
OpenAI Brain

The OpenAI Brain node selects which OpenAI large-language model powers the agent’s reasoning and responses.
This model is invoked whenever the agent “thinks” or generates text during a call.

​
Main Settings

​
Model

Choose the OpenAI model used for this agent. Models vary in speed, cost, and reasoning depth.
The o-series models (o1, o3, o4) are optimized for advanced reasoning tasks, while the GPT-4.x and GPT-5.x families provide strong general performance for real-time voice agents. The dropdown includes:
  • GPT-5.1 and GPT-5.x variants
  • GPT-4.1 and GPT-4.x variants
  • o-series models (o1/o3/o4, Mini/Nano variants)
Select the model that best fits your latency and reasoning requirements.

​
Reasoning Effort (GPT-5.1)

Controls how much internal reasoning GPT-5.1 performs. This setting only applies when using a GPT-5.1 model. Options:
  • None (Recommended for Voice) Choosing None allows OpenAI to automatically route reasoning and is generally preferred for real-time conversational latency.
  • Low
  • Medium
  • High

​
OpenAI Realtime Brain

The OpenAI Realtime Brain is a multimodal realtime model that handles hearing, thinking, and speaking within a single node.
When selected on the Start node or an Agent node, it automatically replaces the Hear, Think, and Speak connectors (unless you enable a separate TTS voice from the settings).

​
Main Settings

​
Use Separate Voice (cheaper)

Controls whether or not the realtime model returns audio. Options:
  • No (Use OpenAI built-in voice) — Use OpenAI’s built-in realtime voice. Simpler but slightly more expensive than using a different Voice model.
  • Yes (Use separate TTS node, cheaper) — Run the realtime model in text-only mode and use a separate Voice model for audio. This is usually cheaper and gives you more control over voices.
    • A Speak connector becomes available so you can attach a separate TTS node.
    • Voice and Speed fields are hidden.

​
Voice

Voice used when the model returns built-in audio.
Only shown when Use Separate Voice = No.

​
Model

Select the OpenAI Realtime model to use.

​
Speed

Controls playback speed of generated audio.
Range: 0.25 – 1.5
Only shown when Use Separate Voice = No.