LLM usage is the primary metered consumption type. Input, output, cache-read, and cache-write tokens are converted to a USD cost and debited from the wallet.

Chat

For each message:

1. Reserve - Holds estimated cost for the selected model (not a fixed default model).
2. Generate - The selected provider runs the model.
3. Finalize - Charges actual tokens using the same pricing rules.

Metering ingest

Developer API metering events with event_type: llm.request are rated using their model and token counts. Events that do not represent billable usage do not debit the wallet.

Overage

When grants are exhausted and overage is configured, remainder may be recorded as overage and invoiced through Stripe.