Spend · Billing Guard

Billing Guard

Check your Claude bill against your own records. RepoOps recomputes what your usage should have cost from the tokens it captured locally, then compares that against what Anthropic actually billed you, day by day and model by model. It finds and explains discrepancies. It never caps, blocks, or throttles spend.

Where to find it

  • Localhost: /billing-guard.html?repo=<id>
  • API: GET /api/billing-guard?repo=<id>
  • Keyboard: K then billing guard
  • Sidebar: Telemetry → Billing Guard

What it does for you

A per-day ledger, one chip per billed day.Each day reads as matches your records (within the tolerance band you set), doesn't match, or not checked. A day RepoOps cannot check is marked not checked, never flagged as a problem.
A findings table you can act on.Every finding is a real mismatch: RepoOps had both your records and the bill for that day, and they disagreed by more than your tolerance band. It shows what you expected, what you were billed, the difference, and which of nine detectors fired. One click drafts a dispute email for you to review before you send it.
The scope gap stays visible.Your local records cover the Claude work done in this repo's folder. An Admin key reports billing for your whole Anthropic organization. When other people or projects bill to the same organization, your bill is legitimately higher, so RepoOps under-reports rather than over-reports. It says so on the page. Treat the totals as a floor.

Connect your account

To compare against a real bill, connect an Anthropic account from the tab itself: paste an Admin key (the one that can read your organization's usage and cost). RepoOps checks it with Anthropic before storing it, keeps it on this machine only, and never shows it again. No environment variable to edit. Set the tolerance band next to the reconciliation table; RepoOps measures your typical drift and suggests a band, but never changes it for you. Export an evidence bundle for finance, in the same format the CLI produces.

Read more