Skip to main content
Use Cost Events for internal usage that needs more detail than the customer-facing Feature Event, such as artificial intelligence tokens, emails, storage operations, or third-party requests. The Cost Model creation flow includes a provider-neutral Artificial Intelligence tokens template. You can enter rates manually or select an effective-dated provider price catalogue version. Nozle creates one ai_tokens meter and separate exact rules for each configured token category. The supported Version One categories are input, cached_input, cache_write, output, and reasoning. Provider catalogue references are retained on the Cost Rule so negotiated overrides remain traceable to their base prices. Unknown provider, model, or token-category combinations enter the Uncosted queue instead of silently becoming zero-cost usage.
At least one attribution field is required:
  • parent_transaction_id links the cost to an existing or future Feature Event in the same organization.
  • external_customer_id records a customer-direct cost that is not linked to a Feature.
  • Both may be provided; asynchronous processing verifies that they resolve to the same customer.
The endpoint requires a secret key. It returns 202 Accepted after Kafka durably acknowledges the message; cost calculation and analytics happen asynchronously. If cost_event_id is omitted, Nozle generates and returns a version 7 universally unique identifier. Reuse the same identifier and payload when retrying a Cost Event.

​
Provider wrapper flow

The Node.js and Python OpenAI and Anthropic wrappers can create the parent Feature Event and its detailed Cost Events together. Set costMeterCode in Node.js or cost_meter_code in Python. The wrapper generates the parent transaction_id before delivery, so child Cost Events may safely arrive before the Feature Event. Each non-zero category is emitted separately with canonical properties:
The provider response remains authoritative to the application. Tracking failures never replace a successful provider response.