MRR and churn belong together
The post reports six paid customers, five churned accounts, and about $99 MRR. These are founder-reported figures, not an audit or a stable forecast.
Looking only at the remaining MRR would hide the retention problem. A small number of payments is not proof of a durable product or personal income.
Ask why customers leave
A creator-data tool needs a recurring job that users cannot easily replace. Interview a churned customer about the last useful moment, the missing result, and the reason the subscription stopped.
Turn the answers into one revised outcome and one reactivation offer. Do not respond to churn by adding a larger model list or unlimited automation.
Use a bounded retention test
Keep a redacted sample, a fixed workflow, and a written success rule. Count model calls, retries, manual repair, support time, and refunds alongside the subscription price.
A small reactivation test can be stopped after a defined number of conversations. If the result is not useful enough to retain, close the experiment before spending more.
Where APIToken fits
APIToken can keep model requests, failures, usage, and budget under an independent project key while the retention workflow is tested.
There is no evidence of APIToken usage in the source. The 83% figure remains a self-reported churn result, not a universal benchmark or an income promise. Keep the customer interview notes and the stop decision with the same experiment record today.