Back to Projects
Axis
Payment authorization and policy engine for autonomous AI agents.
nextjsnodejstypescript
The Problem
AI agents making autonomous financial transactions have no governance layer.. no spending limits, no merchant restrictions, no audit trail, and real risk of runaway or malicious spend.
The Solution
Built a policy engine between agents and payments: scoped wallets, per-transaction caps, period budgets, merchant allowlists, and expiry controls, enforced in under 20ms via a 5-step policy pipeline. Semantic merchant matching runs through an in-memory vector engine ("Ähnlich") using cosine similarity. Payments settle through Korapay virtual NGN accounts on a double-entry, immutable ledger. Shipped as a TypeScript SDK (useaxis, v1.0.1).
// what I learned
Enforcing policy checks at sub-20ms forced a tighter architecture — moving merchant matching to an in-memory vector engine instead of a database round-trip was the difference between a usable and unusable latency budget. Building the ledger as double-entry and immutable from day one also saved a rewrite once real transactions started flowing through Korapay.
