Meter

Every token, accounted for.

The Meter is the observability surface of the platform: live token tracking per route, hard caps enforced at the gateway, and spend your finance team can forecast.

The routes ledger.

Every surface of your product is a named route. The Meter shows what each one used, cost, and served, this second.

How routes pick their models
Meter · this week994M tokens · $149.10 · cap $400.00
prod/support-agent412M$61.80
prod/fraud-review298M$44.70
prod/dispatch-planner181M$27.15
prod/weekly-reporting74M$11.10
staging/*29M$4.35

Watch it live from the terminal.

Every request, streamed as it happens. When something misbehaves you see it in seconds, not on next month's bill.

Install the CLI
allocate tail
$ allocate tail
200 · prod/support-agent · gemini-3.5-flash · 288ms · 1,204 tok · $0.0041
200 · prod/fraud-review · deepseek-v4 · 401ms · 2,880 tok · $0.0026
200 · prod/dispatch-planner· dispatch-model/v3 · 312ms · 1,911 tok · $0.0017
402 · staging/scratch · cap reached, request refused

Metered like a utility.

Because that is what inference is.

Live tail

allocate tail streams every request as it happens: route, model, tokens, latency, cost. Wrangler-tail for tokens.

Per-route truth

Spend rolls up by route, not by mystery. You see exactly which surface costs what, and which model serves it.

Hard caps

Budgets are enforced at the gateway, per key and per route. An agent cannot outspend its policy.

Forecast first

Usage is projected before you commit and visible live after. The invoice is a confirmation, never a reveal.

It starts with one dataset.

Book a build session: we scope a live agent, connect one environment, and forecast your usage, in under an hour.