Beta

Agent Spend Firewall

A policy layer between an agent and its wallet: budgets, intent checks, human approval and a receipt for every payment.

The split

LLM writes: Explains blocked payments to the user in plain language
System One decides: Intent match, endpoint legitimacy, anomaly
Code executes: Budgets, allowlists, limits, signing, receipts

Engine

Code enforces budgets and allowlists first; the brain answers 'does this payment match the user's intent?' and 'is this endpoint what it claims to be?'; a person approves anything above the limit; receipts are signed and stored.

Rulebook

What correct means here, and the mistakes we've already caught. It grows one job at a time.

R01Budgets and allowlists are enforced in code before any model is asked.
R02No payment above the per-transaction limit without a person's approval.
R03Every payment gets a receipt: what, why, which decision, who approved.
R04A model can block a payment. It can never raise a limit.

Intake

The agent's wallet (Celo or Base)
Budgets and allowlists
Who approves what, and where (Telegram or Slack)

UNIT: One agent wallet protected, per month
REVIEW LAYER: Weekly review of blocked and approved payments.
DELIVERY: Middleware for your agent, an approval channel, and a receipts view.
TURNAROUND: 3 days

How we find buyers

Agent builders using x402 and Omni402
Celo and Base agent communities
Hackathons