Skip to content

AI that survives contact with billing.

Anyone can call a model API. The job is everything around the call: metering that can't double-charge, refunds when a provider fails, balances that fail closed, and output that never leaks which vendor you're running.

I've built this three times in production: a nineteen-model gateway with a per-model rate card, a metered per-answer AI you can resell, and a faith companion whose memory outlives the session.

The proof

Work that backs this page.

The scope

What a commission includes

  • Provider-agnostic gateway — swap models without touching product code
  • Idempotent credit metering with refund-on-failure
  • Vendor-string sanitization: your users see your product, not your supplier
  • Fallback routing so one provider's outage isn't your outage
  • Cost observability from day one

Straight answers

Asked often.

Which AI providers do you work with?
Whichever serve the task — my production gateway routes nineteen models across eight providers. The architecture keeps you free to change your mind.
Can you add AI to an existing product?
Yes — that's most of the work I take. The integration is designed around your data and billing, not bolted on.
How do you handle AI costs?
A rate card per model, credits metered idempotently, and refunds when a call fails. You'll know your unit economics before launch.

Tell me what you're building. If I'm the wrong person, I'll say so and point you somewhere better.