Using Tools in Playground

As enterprises increasingly adopt agentic workflows, tool calling has become a crucial component of many production LLM applications. However, debugging tool calls can be complex, as errors may occur at various stages of the pipeline. Key questions to consider include:

• Did the LLM have access to the necessary function to respond to the input message?

• If the correct function was available, did the LLM choose to invoke it?

• Was the function called with the correct parameter values?

• Were there ambiguities in the function description or parameter definitions?

These challenges highlight the need for a user-friendly interface for debugging. Once tracing is set up for function calls, the playground automatically populates function definitions and outputs when loading in a span or a dataset, facilitating precise iteration and refinement. Additionally, you can define and modify functions directly in the playground interface, enabling rapid prototyping and iteration before making changes to the codebase.

The "Functions" button displays how many functions the LLM can access.

Click on "Functions" to add, modify or delete the function definitions in a slide-over.

Specify the function selection methodology for the LLM in a drop-down — for example, specify whether functions can be "auto" selected by the LLM or are "required".

View the function calling output in pretty format.

Click into a row to view the function calling output in a full screen view.

Switch to "Raw" JSON output format for more detailed debugging, which provides additional information, such as the tool call ID.