Production readiness checklist#

Move from your first request to an integration that your team can operate. Assign an owner to each item before launch.

Account and provider setup#

Activate the Marketplace subscription and confirm the purchased plan, billing period and key allowances.
Connect your Microsoft Foundry resource or Google Vertex AI project with the permissions in the relevant guide.
Map and test each deployment. Check its reported billing tier and the endpoint operations supported by your provider.
Confirm provider quotas, region and your application's data requirements. Treat a data-boundary setting as a routing control, not a compliance certification.

Create separate keys for the tenants or applications that need independent limits. Document the key-to-customer mapping internally.
Keep keys in server-side secret storage. Never ship them in browser code, mobile bundles, screenshots, Git commits or support tickets.
Use the exact documented authentication header and API base URL. Existing identifiers still use DCAI, including api-key, x-dca-route and DCA_GATEWAY_KEY.
Set a per-key monthly token budget. Verify the plan's per-minute and monthly caps against expected peak usage.
Establish a key-replacement and revocation procedure, including customer communication and an application rollback path.

Test the real endpoints your application uses: chat, streaming, embeddings or provider-specific operations. A successful chat request does not establish support for every endpoint.
Handle 429 with the documented retry-after value and bounded backoff. Avoid retry loops; retries can generate provider costs.
Handle authentication, exhausted budget and unsupported-provider errors separately. Review errors and troubleshooting.
Set application timeouts and limits. Confirm behavior when a stream disconnects and when a deployment is unavailable.
Verify route mappings such as cheap or oss-only. A named route does not establish automatic cross-provider failover or cheapest-model selection.
Run a controlled load check against your agreed capacity and observe both gateway and provider limits.

Review the pricing estimator with expected token mix and active keys. Include provider inference in your separate total-cost estimate.
Check remaining-token headers and the console's tenant usage. Export CSV and reconcile it with the Gateway Marketplace charge and provider invoice.
Confirm how your billing-period dates differ from calendar-month caps. Track active keys, including Default and test keys.
For managed customers, explain the included service, overage rules and separate provider costs before onboarding.

Record the application owner, billing owner, provider administrator and incident contact.
Follow the support guide for sanitized request details. Share timestamps, endpoint, status and request IDs when available; remove secrets and customer payloads.
Confirm support hours, any response commitments, data terms and escalation contacts in your actual agreement. This checklist does not create an SLA.
Recheck the provider guide after a model, region or deployment change.