Creating Playground from Spans

One of the most powerful features is the ability to create a playground directly from existing LLM spans in your traces. When you click Open in Playground from any LLM span, the playground automatically inherits all the original configurations:

  • Model settings: The exact model used in the original request
  • Token configuration: Max/Min tokens, temperature, thinking tokens
  • Images: Image content also preserved
  • Tools: Any tools configurations
  • System prompts: Complete message history and system instructions
  • Input format: The original prompt structure and variables

This eliminates the need to manually recreate complex configurations and allows you to instantly experiment with variations of successful prompts from your traces.