Documentation / Guides / Plans and usage

PLANS AND USAGE

Weighted units, not request counts.

Pricing status: proposed. The allowances below are the approved commercial shape of the product. No Atlas API plan is purchasable yet: there is no live checkout, no metered overage, and no customer charge. Treat them as the model, not as a bill you can incur today.

Two separate subscriptions

The Eventium application plan and the Atlas API plan are different products, bought separately. Application plans cover Nova, Orion and the specialist surfaces, and are counted in chat units and execution credits. Atlas API plans cover external API access and are counted in AAU — Atlas API Units.

Buying Eventium Pro does not grant API Pro, and buying API Max grants no execution credits, specialist-surface access or marketplace rights. You may hold an Atlas API subscription with no paid application plan at all. The two ladders share tier names and nothing else: an application starter entitlement is not an Atlas API subscription and resolves to API Free.

Atlas API allowances

API Free         10,000 AAU / month      2 active keys     proposed
API Pro         250,000 AAU / month     10 active keys     proposed
API Max       2,000,000 AAU / month     25 active keys     proposed
API Enterprise    contracted           100 active keys     per contract

Enterprise has no published allowance; it is agreed per customer. An Enterprise subject with no contracted figure configured is held to a conservative internal ceiling, and that ceiling is not an entitlement to rely on.

Read your own plan, allowance, consumption and remaining balance from the usage endpoint. It is free to call.

curl https://api.eventium.ai/api/commercial/atlas-api/usage -H "X-Api-Key: $EVENTIUM_API_KEY"

The response also carries plan_basis, which states why you are on that plan: entitlement for a paid subscription, legacy_mapped for a recognised pre-ladder identifier, legacy_unknown for an unrecognised one, or no_legacy_source. Every basis but the first yields API Free.

How AAU are counted

AAU are weighted. A key lookup and a recall across five memory Banks are not the same amount of work, and the rate card prices them accordingly. The current card is atlas-aau-v1-draft.

Exact / paginated read        1 per started 4 KiB returned, minimum 1
Write or update               2 per started 4 KiB normalized, minimum 2
Semantic recall               5 per authorized Bank searched
Bounded graph query           5 per authorized Bank scope
Supported resource mutation   1
Health, usage, revoke, delete 0

Two consequences matter in practice. Recall scales with Banks, not results: a query across three Banks costs 15 AAU whether it returns thirty matches or none, so narrow the authorized Bank set rather than the result count. Every started increment is charged: a 4,097-byte write costs 4 AAU, not 2.

What is never charged

Failed requests cost nothing. Malformed input, a denied scope, a throttled request and a platform-side failure are all 0 AAU: you are billed for work performed on your behalf, not for attempts.

Retries of the same operation are not double-charged. Send an Idempotency-Key and a retry replays the original charge. Without one, a retried write is a new logical operation and is charged as such.

Observing and leaving are free by design — reading usage and revoking a key cost nothing, so an exhausted allowance can never trap you. Atlas work performed inside a first-party Eventium feature is accounted to the application plan and is not also billed to your external API allowance.

Rate limits are a separate control

Allowance answers how much you may consume in a month. Rate limits answer how fast — requests per second, burst capacity and concurrent in-flight requests. A request can be refused for pace while allowance remains, and accepted on pace while exhausting allowance. Rate limits exist for abuse prevention and physical capacity. They are engineering defaults, not throughput guarantees, and are not an SLA.

When you run out

Exhausting the monthly allowance returns 429 with ATLAS_API_QUOTA_EXCEEDED and the plan, allowance, usage and rate-card version in details. The allowance resets at the start of the next monthly period (YYYY-MM, UTC).

Overage is off by default, with a $0 additional-spend cap. Nothing you do at the API starts charging you more; enabling overage is an explicit act by the billing owner, with an explicit ceiling. There is no silent postpaid spend.

Distinguish 429 from 503 {"error":"Plan limits not configured"}. The second is not about your consumption — see Troubleshooting.

Who decides what

Eventium is the entitlement and quota authority: every admission decision is made locally and synchronously, and fails closed if it cannot be determined. Stripe is the financial system of record and is never consulted in the request path. Atlas remains authoritative for identity, memory and permitted operations.

Metering is billing input. No amount of remaining allowance grants a capability you do not otherwise hold.