Execution logs

Execution logs give you a way to inspect any functions called by the LLM, including with their arguments and return values.

Execution logs are useful to understand the calls being made by a specific model, as well as the type of arguments and the number of tokens consumed. Each execution is timestamped so you can understand execution time and optimize latency accordingly.

Execution logs work out-of-the-box across any model, so you won't have to add specific instrumentation to your code.

Using Execution logs

Head over to Execution logs in your Toolhouse. In the left hand side column, you will see your past executions, tagged by their own tool call ID as passed by the LLM. Click on one execution to see its details on the right hand side of the page.

What data is available for each execution

For each tool call, you'll see these details:

Data	Description
Tool call ID	The ID of the tool call, as passed by the LLM. This is useful to tell apart two different calls of the same tool within the same inference call.
Execution timestamp	The date and time of when the tool was called. This is computed by Toolhouse
Duration	The duration of the tool exeuction. This is determined by measuring the delta between the time Toolhouse invokes the tool and the time the tool returns its output to Toolhouse.
Input	Input arguments as generated by the LLM
Output	The return value of the tool