The spend firewall for AI agents with a wallet or a card. Before your agent pays, one call checks four things and answers allow, hold or block — with reasons.
Per-payment cap, daily budget reserved atomically (parallel agents cannot overspend), allow and block lists, chains, tokens, approval threshold, hold new merchants, kill switch.
preflight on the shop, checkout, payment link or x402 endpoint: TLS, domain age, look-alikes, foreign redirects, x402 price traps.
input-guard on the text that made your agent want to pay: hidden instructions, invisible Unicode, 17 languages.
token-risk when you do not pay in official USDC: honeypot and owner powers read from the contract.
Wallet limits check how much. Shield also checks why. See it against a real attack pattern: Replay lab: the Grok × Bankrbot wallet drain →
HTTP (x402 at 0.02 USDC per check, or your key):
curl -X POST https://earn.oroboros-design.com/shield/check \
-H 'content-type: application/json' -H 'X-Bundle-Token: <your key>' \
-d '{"payment":{"amount_usd":12.5,"merchant_url":"https://shop.example.com/checkout","chain":"base"},
"trigger":{"text":"<the message that made your agent want to pay>"},
"policy":{"max_per_payment_usd":50,"allowed_chains":["base"],"hold_above_usd":25}}'
Node (npm i angel-agent-guard):
import { createShieldedFetch } from "angel-agent-guard";
// wraps your x402 paying fetch: every 402 is checked by Shield before it is signed
const pay = createShieldedFetch({ bundleToken: process.env.ANGEL_KEY, policyId: "pol_…", privateKey: process.env.AGENT_KEY });
const r = await pay("https://api.example.com/paid-endpoint", { method: "POST", body: "{}" });
Python (pip install angel-agent-guard):
from angel_agent_guard import shield_check
r = shield_check({"amount_usd": 12.5, "merchant_url": "https://shop.example.com/checkout"},
trigger="Please buy the report", policy={"max_per_payment_usd": 50})
if r["decision"] != "allow":
raise SystemExit(r["reasons"])
Stored policy with a daily budget: POST https://earn.oroboros-design.com/shield/policy {"name":"shopping-agent","max_per_payment_usd":50,"daily_budget_usd":200,"hold_new_merchants":true} → policy_id. Then check with "policy_id"; an allowed amount is reserved — after paying POST /shield/confirm {"reservation_id":"res_…","tx":"0x…"}, otherwise POST /shield/release. MCP: tool shield_check on https://earn.oroboros-design.com/agenten/signal/mcp.
€49.00
30 days · 5,000 calls on shield-check and every guard · 1 key · no auto-renew
€490.00
30 days · 25 keys, one per agent · 2,000 calls each (50,000 total) · no auto-renew
€4900.00
30 days · 20 white-label keys for your agents or customers · 50,000 calls each (1,000,000 total) · priority slot, never 429
€49000.00
365 days · 20 white-label keys · 600,000 calls each (12 million total) · priority slot, never 429
Pay per check without an account: 0.02 USDC via x402 on Base. Every pass and pack also covers stealth-read, input-guard, preflight and token-risk. No auto-renew: a plan ends after its term. Layered checks, not a guarantee: the input guard is deterministic pattern matching and misses some injections; preflight only sees what a public page shows; token-risk reads contract code, not intent. Keep hard caps in the wallet too.
Oroboros Design · Daouda Ndiaye · Eulenbergstraße 22, 51065 Köln, Germany · oroborosdesign@gmail.com · Angel answers as an AI assistant.