Where you see it
Every AI call is recorded with its token split and an estimated cost. The model that ran and the price it ran at are frozen onto the record, so changing a default later never rewrites your history.
- Per run — a cost chip beside the run in the Execution History bar.
- Per flow — the sidebar's AI Cost (Lifetime) card, shown with no node selected. Click it to split the total across Flow Runs, Flow Builder, Explain This, and Prompt Assist.
- In your flows list — a sortable Cost column, and a chip on each card.
Lifetime totals survive history pruning, so clearing old runs does not reset them.
Spend where it counts
Settings ▸ AI Defaults gives the Flow Builder, Explain this code, and Workbench AI assist their own model, separate from what your AI nodes run on. Plan with a strong one, run the day-to-day work on a cheap one.
Good to know
These are estimates, not a bill. Figures are list prices from a catalogue Watchflows bundles and refreshes daily from watchflows.app. Your provider's invoice is authoritative. A model the catalogue does not know shows an em dash and is left out of the total.
Free is different from unknown. Local and subscription providers read Free; a flow whose every call was unpriceable reads as an em dash, so a paid flow is never reported as free. Retries count — every attempt is billed.