Priced like infrastructure, not like a research tool

Calls are priced in credits by what they do: a search, a document fetch, similar, browse and why_not cost 1 credit; corroborate, related, authorities, dossier and path 2; assemble_context 3; ground 5; changes 10. A call that runs and fails costs 1 credit, and a refused call — a 401, 402 or 429 — costs nothing. /account, the MCP handshake, health probes, the capability catalog at /describe, the skill and the machine-readable descriptions are free. Every response says what it charged in an x-credits-charged header.

On the free tier, calls stop — it hard-caps, because there is no card on file and a surprise bill is worse than a clear stop. A call that would cost more than the credits you have left is refused with a 429 rather than run, so an expensive call cannot overshoot the allowance. On paid plans you keep serving and usage-based overage applies; your current rate is shown in the console.

It is how far back that tier may query. Free reaches 30 days, Build a year, Scale five years, and Archive the full index with no date floor. It bounds the date range of your filters, not the freshness of results — every plan sees the newest content.

Per account, once. Your plan's rate — 60 credits a minute on Free up to 12,000 on Archive — is counted in a single window for the whole account, however many machines answer your requests. So the published number is the number you get, not a floor with undocumented headroom above it. Exceed it and the call is refused with a 429 whose retry-after says how many seconds are left in that minute; sleep that value rather than a fixed second.

Archive at $799 a month covers 2,000,000 credits, 12,000 credits a minute and the full archive, which handles most workloads that would elsewhere be called enterprise. If you need more than that, or committed capacity, get in touch. If you need the evidence layer inside your own boundary — a national cloud, your VPC, or a disconnected environment — that is the sovereign edition: the same product, deployed and scoped per installation, and priced per installation rather than per credit. It is not on this table on purpose.

Yes, from the console at any time. Changes apply to every live key on the account in one operation, so you never have a window where some keys are on the old plan.

Because the structure is precomputed. Story clusters, origins, the link graph and the corroboration signals are built when the index is built, so a ground call is retrieval over stored structure rather than a research agent spending model inference on every question — it costs infrastructure money, not research-agent money. We also own the index, so no part of our bill is a per-query fee to an upstream provider, and what we spend follows how much the API is used rather than how large the index is. The cost model page sets out the full argument.