Encyclopedia Evalica / Tracing and instrumentation / Tool call

Tool call

/tool kahl/A structured request from a model to execute an external function, such as a database query, API call, or code execution. Tool calls are typically captured as spans within a trace so they can be debugged and measured. (noun)

A failed tool call caused the agent to fall back to a generic answer.

Customer example

Graphite evaluates its AI code reviewer (Diamond) by curating datasets from real internal PR usage (accepted vs. rejected suggestions) and running custom scorers that measure each tool call in the review workflow, so the team can pick models and ship changes based on measured acceptance-rate impact. Read more

Related Tracing and instrumentation terms

From the docs

Get started with Evals

Braintrust is the AI observability and eval platform for production AI. By connecting evals and observability in one workflow, teams at Notion, Stripe, Zapier, Vercel, and Ramp ship quality AI products at scale.

Start building