Inspect a run

When a flow does something surprising, open the run and look. Workbench's dev tools keep a per-node trace of every execution, including runs from past sessions, share links, and Slack.

1. 1
   Open the dev tools

   On the flow's detail page, toggle Devtools in the subnav. A resizable panel opens along the bottom with tabs: Versions, Runs, Evals, Usage, Quickstart.

2. 2
   Pick a run

   The Runs tab lists executions newest-first — fresh runs from this session merged with persisted history from earlier ones. Guest, Slack, and API runs all land here too, not just your test-pane runs.

3. 3
   Read the trace

   Select a run for two views:

   • Trace — a waterfall of the nodes that fired, synced to the canvas (hover a row, the node highlights). You see each node's status, duration, and start offset from the run's start.
   • I/O — the per-node packets: a two-column inputs / outputs JSON split for each node, plus the exact inputs the run received and, on a failure, the error message and server response.

The Runs view is read-only: it's for understanding what happened. Re-run from the test pane or the API. The Usage tab adds spend over time, so you can see what a flow is costing as traffic grows.

Run budgets

A custom-mode flow can carry a run budget — a cost bound (dollars per run), a latency bound (milliseconds), or both, set in the flow's Details dialog on the canvas. Runs that exceed a bound show that stat in amber in the Runs list, with the budget in the hover text. It's a flag: runs never abort on budget, and nothing changes about how the flow executes. Use it to make "this flow got slower" or "this flow got expensive" visible the day it happens instead of on the invoice.