Execution graph
The span-level trace graph of an agent run — every step, its tokens, and its cost.
The execution graph renders a single agent run as its OTel gen_ai span tree — agent invocation, model calls, tool calls — with cost attached to every step. This page outlines the section; it will be expanded as the view evolves.
What it shows
The full span tree of a run (invoke_agent, generate_content, execute_tool), and for each span: duration, model, token usage, estimated cost, and its reconciliation state once invoice data has been joined.
Questions it answers
What did this run actually do? Which step burned the tokens? Why was this particular call expensive — model choice, prompt size, retries?
Main interactions
Expand and collapse branches of the tree, inspect a span's attributes, and jump from a span to its cost lineage in reconciliation. Runs are reached by drilling in from the fleet view.