I built an authorization layer for AI agents. Here's why it had to be non-custodial.
rust
dev.to
A few months ago I was prototyping an agent that needed to make purchases -- API credits, small SaaS subscriptions, that kind of thing. I started wiring it up to a payment method and immediately hit a wall I couldn't shake: there was no layer between the agent and the funds. No "is this actually okay right now?" check. Just a key with full access. I looked for something that handled agent-specific authorization -- not just API rate limiting, but actual financial authorization with per-agent lim