Going to production

• Use separate keys for development, staging and production.
• Keep keys in environment variables on your server, never in client code.
• If a key leaks, revoke it in the console and create a new one.

The OpenAI SDK retries 429 and server errors for you and waits for Retry-After. Set a retry count and a timeout:

client = OpenAI(
    base_url="https://api.sweetrouter.com/v1",
    api_key=os.environ["SWEETROUTER_API_KEY"],
    max_retries=3,   # retries 429 and 5xx with backoff, honours Retry-After
    timeout=60,      # seconds
)

Spread traffic evenly and handle 429 by waiting and retrying. Expecting a lot of traffic? Tell us your peak requests per second before launch and we'll raise your limits. See Limits.

Requests fail with 402 insufficient_balance when your balance runs out. Check it on the Billing page and top up ahead of busy periods. For invoicing or larger volumes, contact us.

Each response includes sweetrouter.job_id and an x-request-id header. Log both, and include them when you contact us. You can also send the user field with your own end-user id to match calls to users in your logs.

• Production key stored as a server environment variable.
• Retries and a timeout set in your client.
• History trimming in place for long chats.
• Errors shown to users in a friendly way.
• Enough balance for your expected traffic.