Skip to main content

Chat Template Kwargs

chat_template_kwargs is part of a prompt's llm_config. It contains provider-specific chat template options that Agenta passes through unchanged.

To configure it in the UI, see Chat Template Kwargs in the Playground.

When you invoke through Agenta​

When you invoke a deployed prompt through Agenta, Agenta applies the saved chat_template_kwargs for the selected model. You do not need to add anything to the invoke request.

If the prompt falls back to another model, Agenta uses the fallback model's own chat_template_kwargs.

When you fetch the configuration​

When you fetch a prompt configuration and call a provider from your own application, Agenta returns chat_template_kwargs under llm_config.

Example fetched configuration:

{
  "prompt": {
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Answer this: {{question}}"}
    ],
    "llm_config": {
      "model": "qwen/qwen3",
      "temperature": 0.2,
      "chat_template_kwargs": {
        "enable_thinking": false
      }
    },
    "template_format": "curly"
  }
}

Pass this field only to provider clients that support it. If chat_template_kwargs is null or missing, there are no chat template options to forward.

Agenta does not format prompt variables inside chat_template_kwargs; its values are provider options, not prompt text.