Legacy telecom billing was built for minutes and messages. It can't meter AI. Pulse treats AI usage as a first-class billable, rated through the same pipeline as voice and messaging.
Operators launching AI voice and agents are metering usage they can't actually bill, a silent revenue leak. Pulse exists because its first operator hit exactly this wall.
Rate the real cost drivers: LLM input, output, and cache tokens; speech-to-text seconds; text-to-speech characters. Each is priced on its own terms: per-model, per-operation, token tiers, minimums, monthly inclusives.
The many atoms of a single AI call (LLM turns, STT, TTS, the voice leg) roll up under one session into a single invoice line, with full drill-down.
Every rated event captures both what you charge and what the provider costs you, so profitability per AI product is visible, not inferred.
Row-level de-duplication (each event has a unique external id), a corrections pipeline for signed adjustments, and fail-closed handling: unpriced usage is quarantined for review, never silently billed at zero, never guessed.
High-throughput, batched, partitioned ingestion with headroom for the bursty token volumes that break row-per-event systems.
New models map to explicit rate items. A model version is a new priced item, never a silent re-rate. Usage thresholds and overage alerting surface a spike before it becomes a surprise.
Modern, multi-tenant billing for telecom, VoIP, and AI-powered communications providers, with event-level AI usage rating and deep telecom tax built in.
