Explain

quantum$0.003

computing

summarize

this

document$0.005

write

code

parse

returns$0.009

$0.0012

$0.0007

$0.0091

$0.0026

$0.0034

$0.0044

$0.0158

METER

What Meter is

Your AI spend has a treasurer. It never sleeps.

Meter predicts inference costs before they happen, enforces budgets per feature, and autonomously tops up provider credits — so you don't have to worry about running out at 3am.

Live · every call priced and attributed

Meter every call.

Change one base URL. Every AI call now passes through Meter — priced, attributed, and written to the ledger with a measured p50 overhead of 1.49ms on loopback.

✓ Attribution recorded on every row

Pay while you sleep.

When the provider balance runs low, the Treasurer agent charges a pre-approved Prava mandate and tops up the wallet before anything fails. You get the iMessage in the morning.

✓ Auto-topup via Prava mandate

Cut spend anomalies.

Named after the fuse box in a house. When one feature's spend rate triples against its own trailing hour, the Circuit Breaker cuts it off. Everything else keeps flowing. Leaked keys die immediately.

✓ Per-feature circuit breaker

01/03

How to use it

Three steps. All of it real.

Scroll through the process. The code updates in real-time on the right. No mocks, no illustrations. Install takes about a minute.

Point at Meter

Change one base URL.

Your app normally talks straight to api.openai.com. Point it at Meter instead. Now every request arrives at our door — we look at it, price it, write it down, and forward it on. Anthropic works the same way.

✓ Attribution recorded on every row

Budgets as code

Set ceilings per project and per feature.

When a limit is hit, the refusal names the exact scope in X-Meter-Budget-Scope, so a developer reading a 429 knows which line to raise. Two ways in, one place it lands.

✓ Rolling 24h · settled + in-flight holds

Close the loop

Price the outcome, not the token.

Meter cannot know whether a support ticket was resolved — so you tell it. Post an outcome against a trace_id and every call in that trace rolls up behind it. The ledger stops being a cost report and starts being a margin report.

✓ Cost per resolved outcome, not per token

.env

# the only line that changes
OPENAI_BASE_URL=https://meter.acme.dev/v1
OPENAI_API_KEY=mtr_platformeng_e7c9a1b2

ANTHROPIC_BASE_URL=https://meter.acme.dev/v1
ANTHROPIC_API_KEY=mtr_platformeng_e7c9a1b2

meter.yaml

projects:
  - id: acme-app
    ceiling_usd_day: 200
    features:
      - id: summarize
        ceiling_usd_day: 50
      - id: chat
        ceiling_usd_day: 120

annotate.sh

# one resolved ticket, a dozen AI calls, one number
curl -X POST https://meter.acme.dev/v1/annotate \
  -H "Authorization: Bearer mtr_platformeng_e7c9a1b2" \
  -d '{"trace_id":"tkt_9812","outcome":"resolved","value_usd":40}'

# -> {"cost_usd": 0.184, "request_count": 12, "margin_usd": 39.816}

Stop watching charts.
Start paying bills.

Meter is live for select teams. Point your first request at us today and see exactly where the money goes.

Open dashboard →