FinOps for agentic AI

Every token,
accounted for.

Claude Code makes your team 10× faster — and your Anthropic bill a black box. Meter is the drop-in proxy that meters every token, attributes it to a dev, repo and task, and auto-optimizes the spend before the invoice lands.

one env var to install zero signup to start your keys never leave
acme · live spend metering
PROJECTED / MO
$18,267
SAVED BY ROUTING
$11,040
41% under all-Opus baseline
METERING IN PRODUCTION
990tasks attributed
8repos tracked
10seats metered
350calls auto-routed
the affordability crisis

Your bill 6×'d last quarter.
Nobody can say why.

Finance sees one giant Anthropic line item. Eng sees a productivity miracle. Between them is a void where a $0.40 task and a $40 runaway loop look exactly the same — until the invoice lands and someone threatens to claw back the budget.

  • No per-dev, per-repo, or per-task attribution
  • Runaway agent loops burn hundreds before anyone notices
  • Everything runs on the priciest model "to be safe"
ANTHROPIC SPEND · UNMANAGED
Q1Q2+612%
how meter works

A meter, a ledger, and a governor —
on one env var.

01 · Meter

Drop Meter on ANTHROPIC_BASE_URL. It pass-throughs every Claude Code request and records exact token counts, model, and latency. Your API key never leaves your machine.

02 · Attribute

The tagger infers repo, branch, dev and task from each request — so finance gets a real cost ledger and eng leads see a $0.40 edit apart from a $40 loop.

03 · Optimize

Trivial edits auto-downgrade to Haiku. Budgets alert before they blow. Runaway loops get killed. The bill becomes a managed line item.

$0 · no signup

See your last 7 days of Claude Code spend by repo.

The free local CLI reads your existing Claude Code logs and prints a per-repo, per-day cost breakdown. No proxy, no account — just the receipts. The shareable leaderboard is on the house.

meter — bash
$ npx meter-cli last 7d

  Claude Code spend · last 7 days · acme
  ┌──────────────────┬─────────┬──────────┐
  │ repo             │  tasks  │     cost │
  ├──────────────────┼─────────┼──────────┤
  │ payments-api     │     94  │  $418.20 │
  │ web-app          │     76  │  $351.07 │
  │ ml-pipeline      │     61  │  $288.44 │
  │ data-warehouse   │     38  │  $163.91 │
  │ infra-terraform  │     22  │  $ 71.55 │
  └──────────────────┴─────────┴──────────┘
  total  $1,293.17   ▲ 38% vs prior 7d
  routing would save  $402  (31%)

$ meter share        # → leaderboard.png
the control plane

From cost visibility to full FinOps governance.

Real-time cost ledger

Every token priced the moment it’s spent, attributed to a dev, repo, branch and task.

Auto-routing engine

Downgrade trivial edits to Haiku automatically — typically 30–50% off with no quality hit.

Runaway detection

Anomalous loops are flagged and killed via webhook before they cost you hundreds.

Budgets & alerts

Per-team and per-repo budgets with Slack/email alerts before the line gets crossed.

Keys never leave

A local pass-through. Your Anthropic key stays on your machine; only aggregates ship.

Policy & allow-lists

Model allow-lists, governance, and audit logs — the Vanta layer for agent spend.

stop guessing

Turn agentic AI spend from a scary black box
into a managed line item.

free local CLI · $20 / seat / mo for teams · enterprise FinOps on request