Automatic usage tracking
Install the SDK, call init() once, and every AI call is tracked. OpenTelemetry instrumentation captures usage without touching your request path.
- TypeScript SDK (@model-meter/sdk) with one-call init() setup
- Auto-tracking via OpenTelemetry gen_ai.* spans from the Vercel AI SDK — zero code changes
- generateText() and streamText() wrappers for per-operation labeling
- Graceful failure — SDK errors never surface to your primary AI call
- Asynchronous telemetry — events buffered and flushed every 5s or 50 events
Accurate cost attribution
Every event carries project, model, provider, and environment labels, priced with decimal precision against effective-dated pricing records.
- Batch ingestion API with per-event validation
- Idempotent ingestion — retries cannot inflate usage or cost
- Decimal-precision cost engine — no floating-point rounding errors
- Effective-dated model pricing — price changes do not rewrite historical records
- Cost provenance: estimated, provider-reported, or reconciled — clearly labeled
Live spend dashboard
Spend moves in near real-time — see which project, feature, or model is driving cost before the invoice arrives.
- Spend overview with time-series trends
- Provider, model, project, and environment breakdowns
- Event explorer with a filterable request list
- Projects management
Workspaces and access control
Organize spend by workspace and team, with tenant isolation enforced at the database layer.
- Multi-workspace data model with PostgreSQL Row-Level Security
- Workspace roles: Owner, Admin, Engineer, Finance, Viewer
- Key management — create and revoke ingestion keys in the dashboard
- Member invitations
Budgets, alerts, and reconciliation
Set spend limits per workspace or project, get alerted before limits are crossed, and reconcile estimates against provider invoices.
- Workspace and project-level budgets with daily and monthly periods
- Threshold alerts — notified before limits are exceeded
- Provider billing import for invoice reconciliation
- Request-level estimates vs. provider-reported side-by-side
- Cost-to-quality evaluation — compare models on equivalent datasets
Gateway adapters
Collect usage from LLM gateways when SDK instrumentation isn't an option — adapters normalize gateway telemetry into Model Meter's canonical event format.
- LiteLLM — callback-based, per-request cost and token data
- Portkey — webhook-based usage metadata
- Ingestion health — last-seen timestamps and rejected event counts
