How usage is measured and priced

Each AI request produces a record: provider, model, token counts (input/output/cached), tool calls, compute time, and timestamps. Prompts and source code are never sent or stored for billing.

Pricing

The platform resolves a versioned pricing rule per usage category at request time — plan-specific rules first, then provider/model-specific, then wildcards. Historical records keep the rule that applied when they ran.

Local inference

Records with mode LOCAL (models on your own hardware) appear in analytics but are never billed.