USDC policy layer for autonomous work

Agent Spending Vault

Give an AI agent a real USDC budget on Arc. It can pay approved services, queue risky actions for approval, and leave a readable onchain settlement trail.

Policy decision
agent requests 0.42 USDC
recipient Verified Data API
rule max 2.00 USDC per action
result executed or queued
What it is

A control plane for AI agents with money

The vault sits between the user, the agent, and service providers. Safe payments execute immediately. Risky actions become approval requests.

01

Fund

Deposit USDC into an agent-owned work budget on Arc Testnet.

02

Constrain

Set recipients, per-action limits, daily caps, and pause controls.

03

Settle

Every payment gets a metadata hash, policy decision, and transaction trail.

1 User sets budget

25 USDC, approved vendors, 2 USDC max per action.

2 Agent requests spend

The agent tries to buy data, verification, or compute.

3 Vault enforces policy

Allowed payments execute. Risky ones wait for human approval.

4 Trail stays readable

Users see who was paid, why, and how the policy decided.