Can CueCrux tell me what my agents cost?

Yes, and the operative word is attribution, not estimation.

Most teams can name their total agent spend and nothing else. The interesting questions (which agent, doing what, on whose behalf) get answered with sampling and guesswork. CueCrux answers them with receipts. Every unit of spend resolves down a single chain: fleet, to agent, to verb, to the signed receipt that recorded the action. Nothing along that chain is estimated; if a number appears in the attribution, a receipt backs it.

The billing plumbing follows the same rule. Local work is free and stays free, including the local dense retrieval lane. Metered work (hosted dense rerank, extraction, attestations) is priced in credits at roughly £0.01 per credit, and every credit spend mints a signed receipt at the moment it happens. Your bill is not a monthly aggregate you take on faith; it is a chain of receipts you can audit action by action, and dispute at the level of a single verb.

Token discipline keeps the underlying costs sane. Every retrieval call carries a hard token_budget, and budgeted retrieval returns far fewer tokens than naive top-K, which shows up directly in what your model provider charges you.

Fleet-wide attribution ships in the Governance tier. The receipts underneath it are minted by the free daemon from day one, so the evidence exists before you pay for the roll-up.