Cost Accounting

Every span records token counts — get per-model, per-session, and per-device cost breakdowns at any time granularity.

Because token counts are captured on every llm_call span automatically, cost accounting needs no separate instrumentation — it falls out of tracing that’s already running.

• Use service_name, environment, device_id, and session IDs (set at `init()` time or per-call) to slice cost by team, deployment, or customer.
• View aggregated token usage by model, provider, session, and device from the dashboard.
• Combine with per-model access grants to prevent a stray integration from silently running an expensive model.