Keep a free API experiment free.

Before sending the first request, establish what will stop your experiment from spending more than you intend.

Choose the exact model first

A platform can offer both free and paid models. Read the price and quota for the model ID you will actually call. A free web interface does not prove the API is free.

Find the hard stop

Determine whether the free plan rejects requests after the allowance, or whether billing continues. A budget alert is not necessarily a spending cap. Check the provider’s documentation for the exact behaviour.

Bound retries and concurrency

Put a maximum number of attempts on retries, use backoff for rate-limit responses, and cap concurrent requests. A loop that retries indefinitely can consume your allowance without producing useful output.

Protect the key

Keep API keys on your server or in a local environment variable. Do not put secret keys in public JavaScript, source control, screenshots or analytics. Rotate a key if it leaks.

Run a small test

Send one request, check the usage dashboard, and confirm the observed usage matches the documented unit. Then set an explicit request or token budget in your application.

Check before launch

• Model and endpoint are eligible for the free tier.
• Output length and request count are bounded.
• Retry limits and timeouts are configured.
• Paid upgrades and fallback models are intentional.
• Usage is monitored and the key can be revoked quickly.

This is an operational checklist. Provider documentation controls actual billing and limits.