Flat pricing, metered simply

Every plan is a flat rate on the calls you send — never a percentage of your spend, and never a cut of your savings. What Metergraph finds is yours to keep.

Free

For trying Metergraph on a real service and seeing your spend clearly for the first time.

$0
100K calls / month included
SDK trace capture — one wrap call, ~5-minute setup
Per-function spend attribution
Console: spend, tokens, latency, model mix
All product features: evals, recommendations, alerts, canaries, datasets, and MCP access
Scale

For organizations with traffic that can't leave their environment.

Custom
Volume pricing on the same per-call meter
Everything in Growth, with negotiated allowances and deployment support
VPC or on-prem deployment
SSO & role-based access control
Managed optimization: our team runs the reports and ships the PRs with you

Building something now? We're onboarding a small number of design partners and working closely with each one — book a call to see if it's a fit.

Questions

What counts as a call?

One captured LLM request/response pair, recorded by the SDK or exported from your gateway. Spans grouped under a trace each count once.

Why not a percentage of savings?

Because it puts our incentives against yours: percentage pricing rewards vendors for big one-time cuts, not for keeping quality up as models and traffic drift. Flat metering means the recommendations only have to be right.

Does Metergraph sit in my request path?

No. Collection is shadow-only and asynchronous — Metergraph never holds live inference credentials and can't fail your product's calls. Optimizations ship through your normal deploy process, validated first.

Our traffic can't leave our environment.

That's what Scale is for: start with staging-only ingestion, then deploy Metergraph into your VPC or on-prem so production traces never leave your boundary.

Start metering today

One SDK call. Your traces do the rest.
Get started Book a call

Or leave your email and we'll reach out. No newsletter, no sharing.