Skip to content

Usage audit (local CSV)

Recompute your API spend from your own usage exports using official tier, cache, and long-context rates — then reconcile it against your cost file, model switch scenarios, and export a migration plan. Every number carries a precision label: exact estimate range

Runs entirely in your browser

Files are parsed locally with the File API. There is no upload endpoint — open your network tab while you work; nothing containing your data is sent anywhere. Analytics record only format names, row counts, and size buckets.

1 · Usage file

Anthropic usage CSV (Admin API cookbook columns) or OpenAI activity export. Or load a synthetic sample:

2 · Cost file (for reconciliation)

Provider cost/line-item CSV (amount_usd or raw-API amount). Unknown currency units refuse reconciliation; credits/refunds (negative amounts) are netted.

What this can and cannot compute

Exact

  • Token split into uncached / cache-write / cache-read buckets (official export fields)
  • Service-tier rate selection (standard/batch/flex/fast where the file carries the tier)
  • Flat-priced models' full recompute, and cost-line matching (MATCHED within one cent)

Estimate or range — never silently exact

  • Cache writes merged 5m/1h in exports → cost shown as a range between the two TTL rates
  • Long-context banding needs per-request sizes; exports are aggregated (buckets, not requests) → tiered models show below/above-threshold ranges
  • Missing tier column → assumed standard, labeled; unknown currency units → reconciliation refused, never guessed

Format support follows the official export shapes documented in our methodology. Synthetic samples are generated from those column specs — they are not real data.