OpenAI & Anthropic auto-capture
Wrap your client or route through the LLM proxy. Tokens — cached, reasoning, audio and image included — land on the event, streaming too.
- wrapOpenAI · wrapAnthropic · LLM proxy
- Per-token-type cost
Every customer action lands as an event, and its cost is rated as it lands — no nightly join, no COGS spreadsheet, no waiting for the provider invoice.
Five steps. One shape. The same row every other layer of the platform reads.
Revenue and cost from the same event. Margin stops being a month-end report — it becomes a number you can query any second.
What runs on every event to keep the ledger honest — the ingest contract, the cost rule, the event pipeline, the uncosted queue.
Replays of evt_5d3... return the same row. No double-bill, no skew, no nightly dedupe job.
Auto-capture for OpenAI and Anthropic, your own rates on a Cost Sheet for every unit-priced service, and shared cost pools for everything you can't pin to one event. One ledger, three intakes, the same math.
Wrap your client or route through the LLM proxy. Tokens — cached, reasoning, audio and image included — land on the event, streaming too.
Your rates for the units you're billed on — AI tokens, Lambda GB-seconds, egress GB, emails sent — applied as each event is rated. Enter your negotiated rates, or register your own internal SKU.
Anything you can't pin to one event — servers, SaaS seats, data vendors. Create a Shared Cost Pool for the period and Nozle allocates it across eligible customers.
Nozle is seven layers on one ledger. The events you send here are the events every other layer reads.
Wire the SDK once and every action is metered and costed as it lands — no nightly join, no COGS spreadsheet.