Back to blog
observabilitygovernance

Non-custodial spend tracking for AI agents

5 min read

The first time a fleet of agents runs up a real bill, most teams reach for the obvious tool: the wallet. And the wallet lets them down. It shows a flat list of outgoing transfers — amounts and timestamps — with no answer to the only questions that matter: which agent paid, for what, and was it worth it? That gap is what agent spend observability exists to close.

Why a wallet isn't enough

A blockchain wallet is a ledger of value moving, not a record of why. For agentic payments it's missing almost everything you'd want:

  • No mapping from a payment to the agent that made it.
  • No mapping to the domain or resource it paid for.
  • No sense of how many paywalls were hit versus actually paid.
  • No latency or failure context — you can't tell a healthy payment from a broken one.

You can reconstruct some of this by hand for a few transactions. You cannot do it across a fleet of agents making thousands of x402 payments a day.

What observability actually means here

Good agent spend observability turns that flat ledger into an operational picture:

  • Spend over time, broken down per agent and per destination domain — so you know where the money goes, not just that it left.
  • Paywalls hit vs. payments made, a ratio a wallet can never show you, which tells you how often your agents are running into priced resources.
  • Per-resource detail — which endpoints cost the most, which agent is the heavy spender.
  • Reliability signals — latency, error rates, and the paid-but-failed cases sellers never surface to the buyer.

The point is to move from "we spent $412 this week" to "agent A spent 60% of that hammering one domain's premium endpoint, and 8% of those payments failed" — the difference between a number and a decision.

Why it must be non-custodial

There's a tempting shortcut: route agent payments through a service that holds the funds and therefore sees everything. Don't. The moment a third party takes custody of your agents' money, you've added the single largest risk in the system — funds that can be frozen, lost, or misused by someone other than you.

Non-custodial observability watches the payment traffic without ever holding the money. It observes the 402 challenges your agents hit and the payments they make from their own wallets; it never signs, holds, or settles funds. You get the full picture and keep sole control of the value. Visibility and custody are separable — and they should be separated.

From observability to governance

Seeing spend is step one. The reason to see it is to govern it:

  • Alerts when spend crosses a threshold, or when a spike suggests a looping agent.
  • Failure alerts when payments start going through but resources still don't.
  • Per-agent attribution so you can throttle or investigate a single misbehaving agent.

That's the arc: observe, understand, control — without ever taking custody of a cent.

402.report is the independent, non-custodial control plane for the buy side of x402: spend tracking, payment forensics, and alerts for teams running autonomous agents. Join the waitlist to see what your agents pay before the bill does the talking.