Thoughts on AI cost management and building sustainable products.
OpenAI and Anthropic have no native per-user cost controls. Here's the architecture that actually works — Redis counters, edge function intercepts, and fire-and-forget logging — plus how to handle streaming.
Read more →Request-count limits miss the point. Spend-based limits are what actually protect your margins. Here's how to implement them properly without building your own Redis infrastructure.
Read more →A practical breakdown of what GPT-4o, GPT-4.1, Claude Sonnet, and Claude Haiku actually cost at scale — and how to pick the right model for your app without getting surprised by the bill.
Read more →We built Nasca because we kept running into the same problem: one power user could burn through your entire monthly OpenAI budget overnight. Here's how we solved it.
Read more →