Credits and billing

Caps and confirmation#

A spending cap is optional. With --max-cost USD the run starts only if the quote's maximum fits under it. Without one, your workspace credit bounds the run. At a terminal, the CLI shows the quote and asks before it charges anything.

An API key can carry its own spend ceiling across everything it starts. Set it when you create the key, or change it later with --spend-ceiling USD|none. A per-run cap still applies to each run. The API writes money as strings of integer micro-USD ("1000000" is $1).

Holds, charges and stops#

• A hold reduces available credit while work is in flight. It isn't a charge. Results show actual EvalRouter spend, and billing may finish after a score appears.
• A run that reaches its cap stops new work and keeps its completed results. Cancelling stops new work but can't retract a call already sent.
• EvalRouter's own infrastructure can end an attempt. Examples are a lost hosted or remote owner, a lost sandbox, an upstream provider failure, or EvalRouter stopping the run mid-attempt. EvalRouter doesn't charge that attempt's model calls, and its retry pays only for its own calls.
• Your provider bills calls to your own provider endpoint or agent. These calls carry no EvalRouter fee, and EvalRouter's cap doesn't limit them. Check the external-cost notice before you start.

Settled charges#

The billing page's Settled charges is everything settled against the wallet for evaluations. It has two parts:

• Usage: metered model inference and compute.
• Platform fees: EvalRouter's per-evaluation fee, net of platform fee credits. The quote shows it, and the ledger charges it as separate entries.

You pay credit purchase fees at checkout, and Settled charges excludes them. The wallet endpoint reports the same split:

Cost forecasts appear only where your deployment offers them. An absent forecast isn't a zero-cost estimate.