Production Checklist

Before launch

• Create separate keys for development, staging, and production.
• Create keys in the correct personal or company account context.
• Keep all keys in a secret manager and test key rotation.
• Fetch the model list and verify every required capability.
• Set request, connection, and total operation timeouts.
• Add exponential backoff with jitter for temporary failures.
• Add an idempotency key to retryable completion requests.
• Add a stable session ID per conversation.
• Handle streamed error frames as well as HTTP error statuses.
• Monitor both h5 and weekly usage windows.
• Alert before quota reaches 100 percent.
• Enforce your own per-user limits before calling the API.
• Validate tool calls and structured output.
• Redact secrets and sensitive prompt content from logs.
• Store the request ID with each failure report.

Test failure cases

Test invalid and revoked keys, unknown models, empty messages, quota exhaustion, rate limiting, timeouts, a connection drop during streaming, duplicate idempotency keys, malformed tool arguments, oversized input, and a model without the requested capability. Confirm that your product gives the user a clear next action for each case.