AI Agent Cost Visibility

A coding agent that loops can burn budget in an afternoon. The invoice arrives at the end of the month, long after the run that caused it. Cost visibility for AI agents means knowing what each run costs while the run is governed, and having a stop that fires before the invoice does.

Cost per run

Modulo records cost at the run, which is the unit of work you actually reason about. The run detail shows a total run cost, a breakdown by named cost components, and the model cost per node. The per-node view carries a deliberate caveat: model and token cost per node never equals the run total, because sandbox infrastructure is an organization-level component. See Pipelines for the run model.

Aggregates roll up from the same facts: a total across runs, an average cost per run, and a cost per successful run, grouped by day, week, month, or year and by team.

Budget controls

Admins set daily spend limits at the organization level and the team level. The check is atomic, and a run that would exceed a limit is rejected before it starts. See Cost Management.

Above the daily limit sit hard ceilings. An organization lifetime ceiling stops brand-new runs before any billable cost. A per-run ceiling caps a single run, enforced at the run's terminal ledger. A run that overshoots is billed to completion and then stopped from accruing further. A breach terminates the run with a budget_exceeded or cost_ceiling_exceeded status, so the stop is visible in the record rather than inferred from a bill.

Anomalies and reports

Modulo compares daily organization spend against a rolling 7-day average and flags a day as anomalous when spend exceeds twice that average. Anomalies appear for an admin to review or dismiss.

Cost data exports as CSV, and scheduled reports deliver daily, weekly, or monthly to named recipients.

Where this sits

Cost visibility is part of governance, not a separate dashboard bolted on. The team that owns the pipeline owns the spend, the same run that produced the cost carries its snapshot and audit events, and the ceiling that stops the run is the one the run's own ledger enforces.

For the other half of the spend question - choosing and routing models so the same work costs less - see Own your LLM spend.

Where to go next

• Cost Management: limits, exports, and scheduled reports
• Model Backends: where model routing is configured
• Govern Coding Agents Across Teams: per-team attribution in context