Skip to main content

LLM Tab

Corrected 19 September 2026: there is no LLM tab. The model is chosen in the agent’s settings (web: Agents > Configure > Settings, or the model menu on each agent’s card; phone: Agents > tap the agent > Edit > Voice > Engines > AI model). The model list below was corrected against docs/user-guide/ (Part 3, section 3.3). The other settings on this page were not re-checked.
The LLM Tab allows you to configure the “brain” of your voice AI agent. You can select the provider and specific model that best fits your latency and intelligence requirements.

Supported LLM Models

  • Anthropic: Claude Haiku 4.5 (the default for new agents), Claude Sonnet 4.6.
  • OpenAI: GPT-5.5, GPT-5.4 Mini, GPT-4o, GPT-4o Mini.
  • Google: Gemini 2.5 Flash.
  • Groq: two Groq-hosted models for very low latency.
Bigger models can sound slower to answer.

Key Configurations

1. Temperature

Controls the “creativity” or “randomness” of the agent’s output.
  • Low (0.0 - 0.3): Highly consistent and factual. Recommended for technical support or bank agents.
  • Medium (0.4 - 0.7): Balanced. Good for general customer service.
  • High (0.8 - 1.0): More varied responses. Good for storytelling or informal chat.

2. Max Tokens

Limits the length of the agent’s response. Note: Voice AI agents should generally have short responses of 50-100 tokens to keep latency low.

3. Context Window

Determines how many previous turns of the conversation the agent remembers. Typically, the last 5-10 turns are sufficient for most tasks.