Measure Everything. Miss Nothing.
Ingest billions of high-cardinality usage events in real time. Aggregate by any dimension. Enforce rate limits and quotas. Power cost tracking and billing from the same event stream.
.png)
Events in. Insights out. In under 100ms.
Send one meter event per AI call. Amberflo ingests it, validates it, aggregates it, and makes it available for dashboards, alerts, and billing—all in under 100 milliseconds.
- Tokens consumed
- API calls
- GPU seconds
- Documents processed
- Custom events
- Image generations
The right aggregation for every metric
Choose how each meter accumulates—from simple sums to distinct user counts and peak-value tracking.
Accumulate the total value of any numeric field
The most common meter type. Use it to track total tokens, total bytes transferred, total GPU-seconds, or any metric where you want the running total over a period.
- Bills customers on total consumption
- Supports partial-period proration
- Works with any numeric event property
- Paired directly with per-unit pricing tiers
Count every event regardless of its value
Use Count when each event itself is the unit—API requests, image generations, inference calls, document uploads. Every event increments the counter by exactly one.
- Track requests, calls, and operations
- Per-call pricing models
- Rate limit enforcement by call volume
- Combine with Sum for dual-metric plans
Capture peak usage within any time window
Max meters record the highest value seen in a window—ideal for tracking peak concurrent connections, peak active users, or high-water storage usage for capacity-based billing.
- Peak concurrency billing models
- High-water-mark storage pricing
- SLA headroom monitoring
- Configurable window granularity
Capture peak usage within any time window
Max meters record the highest value seen in a window—ideal for tracking peak concurrent connections, peak active users, or high-water storage usage for capacity-based billing.
- Peak concurrency billing models
- High-water-mark storage pricing
- SLA headroom monitoring
- Configurable window granularity
Enforce limits before costs spiral
Attach rate limits and quotas to any meter. Amberflo evaluates every incoming event against live usage data and blocks or throttles requests the instant a threshold is hit.
- Per-minute, per-hour, and per-month windows
- Per-customer and per-team enforcement
- Hard block or soft throttle responses
- Quota reset on billing cycle rollover
- Configurable grace period before blocking
Everything a finance-grade metering system needs
Built to handle the reliability, scale, and flexibility demands of AI-native applications.
Sub-100ms End-to-End
From event submission to dashboard update i under 100 milliseconds. No batch jobs, no delayed aggregations—metering that keeps pace with your product.
One Event, Dual Purpose
The same meter event simultaneously powers internal cost dashboards and customer-facing billing. No duplicate instrumentation, no data sync issues.
Idempotent Ingestion
Every event includes an idempotency key. Amberflo deduplicates automatically, so retries and network blips never double-count your usage data.
Custom Ingest Schema
Already have events flowing to your data pipeline? Map your existing event schema to Amberflo meters without changing your instrumentation code.
Real-Time Rate Limiting
Enforce per-customer, per-team, or per-featur rate limits in real time. Prevent runaway costs with hard blocks or graceful throttling.
Meter Health Monitoring
Track ingestion rates, error rates, and latency for every meter. Get alerted on anomalous event volumes before they become billing surprises.
Meter everything. Query anything.
Attach custom dimensions to every event. Filter, group, and aggregate by any combination—without changing your meter definitions.
Custom dimensions. Zero schema migrations.
Send any key-value pair as a dimension on any event. Amberflo indexes them all automatically—no pre-registration, no schema changes, no downtime.
- High-cardinality support
Millions of unique dimension values are indexed and queryable in real time—user IDs, model versions, geographic regions, request types. - Retroactive queries
Add a new dimension to your events today and query historical data by that dimension back to day one—no backfill jobs required. - Dimension-based pricing rules
Apply different rates based on dimension values—charge more for premium models, discount bulk team plans, or offer per-region pricing. - Same dimensions for cost + billing
Every dimension you use for internal cost attribution is also available for customer- facing billing breakdowns—one event, dual purpose.
Metering you can trust at billing time
Send one event per AI call. Amberflo handles attribution, pricing, alerts, and billing from there.
Idempotent Ingestion
Attach an idempotency key to any event. Duplicate submissions— from retries or bugs—are automatically deduplicated.
Late Arrival Handling
Events that arrive late due to network delays are correctly attributed to their original timestamp, not the arrival time.
Historical Backfill
Missed events or data migrations? The backfill API lets you ingest historical events and retroactively update usage and cost records.
Schema Validation
Every event is validated on ingest. Malformed events are rejected with clear error messages—no silent data loss.
.avif)
.png)
.png)
.png)
