FinOps for agentic AI

Your agents spent $36,366 this month. Can you say on what?

Joule is a drop-in proxy that meters, attributes, and auto-optimizes every Claude Code token your team burns — turning an unpredictable bill into a line item you can defend.

Install the free CLI See a live dashboard
One env var to onboard · $735 saved / 14d by routing · Zero signup to start
Engineering · monthly budget live
83%
$36,366 / $44,000
Opus 4.8$13,128
Sonnet 4.6$3,643
Haiku 4.5$200
$1,212
Run-rate / day
184 tasks
Optimized
3.07B
Tokens / 14d
The black box

The invoice is one number. Your spend has a thousand sources.

Per-API-key dashboards tell you the total. They can't tell a $0.40 refactor from a $40 runaway loop, or which repo, branch, or engineer burned the budget — until the bill lands.

5–10×
Typical Claude Code bill growth per quarter after rollout
0
Per-dev, per-repo, per-task attribution in the raw invoice
#1
Reason finance threatens to claw back AI budgets
100×
Cost gap between a trivial edit and a stuck agent loop
What Joule does

Meter it. Attribute it. Optimize it.

Meter every token

A pass-through proxy on ANTHROPIC_BASE_URL counts input, output and cache tokens on every request — no SDK changes, no code in your app.

Attribute to a source

The tagger infers repo, branch, engineer and task from each request, so cost rolls up to a person and a line of work — not an opaque key.

Optimize the spend

Trivial edits auto-route Sonnet → Haiku and runaway loops get killed before they bill, cutting 30–50% with no quality regression flagged.

Shared dashboard

A cost ledger finance and eng read the same way.

Per-seat and per-repo breakdowns, budget gauges, model mix, and a real-time alert feed for runaway loops and budget breaches. Defend the budget with ROI data — then expand it.

  • Per-seat & per-repo cost leaderboards
  • Budget gauges with projected month-end
  • Runaway-loop detection & auto-kill
  • Routing savings, tracked to the dollar
Open the dashboard

Cost by engineer

14d
Priya Nair$2,194
Sofia Ramos$1,939
Elena Petrov$1,803
Tom Becker$1,782
Runaway loop killed 12m
payments-svc burned $41.78 in 38 min — auto-paused.
Routing optimization

The cheapest token is the one you didn't need to spend.

Joule scores each task's complexity and routes it to the smallest model that holds quality. Trivial edits drop to Haiku; deep reasoning stays on Opus. Every downgrade is logged with the dollars it saved, so the savings are auditable — not a vibe.

30–50%
typical spend cut
$735
saved in last 14 days
Model mix · by cost 14d
$17k
metered
Opus 4.877%
Sonnet 4.621%
Haiku 4.51%
184 tasks auto-downgraded
Start free · zero signup

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

A $0 local CLI that reads your existing Claude Code logs and prints a per-repo, per-day cost breakdown. No account, no proxy, no data leaves your machine.

bash — joule cli
$ npx joule scan --last 7d

  JOULE  metered 312 sessions across 7 repos · 7 days

  REPO                 SPEND     SESSIONS   TOP MODEL
  ──────────────────────────────────────────────────
  web-app            $1,284.20      96     Sonnet 4.6
  payments-svc       $  962.55      41     Opus 4.8
  ml-pipeline        $  748.10      33     Opus 4.8
  infra-terraform    $  402.88      58     Sonnet 4.6
  mobile-ios         $  311.40      44     Sonnet 4.6
  ──────────────────────────────────────────────────
  ! 2 runaway loops detected · $70.18 in 2 tasks
  → routing 142 trivial edits to Haiku would save ~$312/wk

  share your leaderboard: joule scan --share

Want shared dashboards, alerts and runaway-loop kills for the team? See team plans →

Pricing

Start free. Pay when it's the team's problem.

Local CLI

$0

For the curious solo dev.

  • Per-repo, per-day breakdown
  • Reads existing Claude Code logs
  • Runaway-loop detection
  • Shareable cost leaderboard
Install CLI
Most popular

Team

$20/seat/mo

For eng teams running fleets.

  • Everything in CLI, plus:
  • Shared dashboards & attribution
  • Budget alerts & auto-kill
  • Routing optimization
  • Slack / webhook anomaly alerts
Start 14-day trial

Enterprise

Custom

FinOps control plane for all agent traffic.

  • Everything in Team, plus:
  • Policy, model allow-lists & audit
  • Usage-based optimization take-rate
  • SSO/SAML governance add-on
  • Self-host / VPC deploy
Talk to us

Put your AI spend on the meter.

Install in one line. See where every joule of agent work goes — then stop paying for the ones that don't.

Install free CLIExplore dashboard