Test tool calls

Debug tool calls directly in prompt playground

Tools are a key part of agentic workflows, but debugging tool use can be tricky:

• Were there any unclear parts in the function description or parameter details?

• Was the function used with the right parameter values?

• If the correct function existed, did the LLM use it?

• Did the LLM have the needed function to reply to the input?

The playground automatically pulls in tools from your span data, making it easy to iterate on tool call prompts, especially if you have traced your function calls.

Configure/add tools when you click the Functions button.

• Functions are calibrated as JSON objects. To add multiple functions, simply populate the list with more JSON objects.

• Configure your Functions Selection. This gives the LLM extra instructions on how/when to decide a function from the list should be invoked.

Here's an example of an LLM's output being a function in Prompt Playground. You can see it gives the function name, as well as the required argument.