Metering, cost and limits
Metering that matches the invoice
Every request is metered by input, output and cache tokens at the provider's real prices, with no prompts stored.
What it is
Every request records input, output, cached and cache-write tokens, cost, latency, time to first byte, modality, finish reason, status, client and source IP. Request and response bodies are never stored as part of metering.
Why you want it
Finance needs numbers that reconcile with the provider's invoice, and security needs to know prompts are not sitting in a database.
How it works
- Money stored as integer nano-dollars, never floats
- 5-minute and 1-hour prompt-cache writes metered and priced separately
- Provider-reported costs (xAI cost ticks) honored; media without usage recorded as unmetered rather than guessed
- Streams that end early are recorded as failures and estimated from their text
- Each response names how it was metered (X-Janus-Token-Accounting)
See it on your own network.
The Community edition is free for up to 25 people. The 30-day Business trial unlocks every Business feature.