Billing & wallet
NexoAI uses a prepaid USD wallet. There is no subscription and no automatic open-ended usage invoice: requests spend existing credit, and service stops when the available wallet or a hard key budget is exhausted.
Rechargez en MAD
- Open the NexoAI dashboard and go to the wallet or top-up view.
- Choose a supported prepaid amount.
- Review the local-rail payment instructions, MAD amount, and wallet credit displayed before confirmation.
- Complete the payment through the supported local rail.
- Confirm that the USD wallet balance updates before starting a large workload.
Top-ups add credit to the same wallet used by every named virtual key on the account. They do not create a separate subscription, reseller balance, or sub-agent account.
Three layers of control
| Layer | Purpose | What happens at the limit |
|---|---|---|
| Wallet balance | Maximum prepaid credit available to the account | Requests that require more spendable credit stop. |
| Key budget | Maximum total spend through one virtual key | That key stops even if the shared wallet still has funds. |
| RPM / TPM | Maximum request and token throughput per minute | Excess traffic is rate-limited until capacity is available. |
These controls work together. A key needs headroom at every layer to serve a request.
RPM and TPM
RPM limits how many requests a key can make in a minute. TPM limits how many input and output tokens it can process over the corresponding window. A workload can hit TPM with only a few large-context requests, or hit RPM with many tiny requests.
Set caps from observed usage:
- Measure normal request size and peak concurrency.
- Allow a deliberate amount of burst capacity.
- Keep the total key budget at the maximum acceptable spend.
- Revisit caps after a model or context-size change.
Retrying can amplify limits
Immediate parallel retries can turn one rate-limited request into a larger incident. Use exponential backoff with jitter and a retry ceiling.
Budget isolation
Create separate keys for workloads that need separate blast radii—for example local-codex, staging-api, and production-worker. A compromised development key can then be revoked without rotating production, and each workload can have an appropriate budget.
Key budgets do not reserve wallet funds. If two keys are active, they still draw from the shared prepaid balance. Monitor the wallet alongside individual key usage.
Reconcile usage
Use the dashboard to compare model usage, key attribution, and wallet movement. For applications with their own metering, record the model ID and returned usage fields per request, but never persist the authorization header.
If a payment or usage entry looks wrong, collect the dashboard transaction reference, timestamp, amount, and affected key name before contacting support. Do not include the actual sk-… credential.
Start strict, then widen
A new integration should begin with a low key budget and conservative throughput. Raise limits after the request path and expected token volume are proven.
See Models & Pricing for the per-million calculation or FAQ for payment and limit troubleshooting.