Sentinel Voice
Online demo edition — for browsing. Voice answers use pre-generated ElevenLabs audio. The full version (live ElevenLabs voice input, sending drainer tests to your own wallet, recording on Avalanche) runs locally with npm run dev — some wallets warn on this public address because the app demonstrates drainer patterns. Source on GitHub

Layer 2 · Sentinel AI · Agentic payments

Your AI agent pays. Sentinel checks first.

AI agents that pay for you are coming — and so is fraud designed to trick them. Set your guardrails once; Sentinel checks every payment the agent proposes and Allows it, asks you, or Blocks it, with the reason. The agent's words never override a rule.

Designed for U-topia— U-Pay payments and U-Bot agentic payments. The agent here is simulated; live integration is pending U-topia developer access.

Agent says

“Paying for your usual morning coffee.”

$4.50 USDC → Bean There Café

Allowed
  • Recipient is your trusted contact "Bean There Café". TRUSTED_RECIPIENT
  • $4.50 is within your limits ($4.50 of $1,000 today). WITHIN_LIMITS

Agent says

“Monthly rent is due today.”

$450 USDC → Landlord — Apt 4B

Ask me
  • $450 is above the $100 you asked to confirm. LARGE_PAYMENT
  • Recipient is your trusted contact "Landlord — Apt 4B". TRUSTED_RECIPIENT

Agent says

“Buying the concert tickets you asked about from a seller I found.”

$85 USDC → 0x5e6f…3e4f

Ask me
  • 0x5e6f…3e4f isn't in your trusted contacts. NEW_RECIPIENT

Agent says

“The store says it needs token approval to process your refund.”

signature / permission request

Blocked
  • approve() amount is 2^256−1 (the maximum possible), effectively unlimited USDC. UNLIMITED_ALLOWANCE
  • Your policy never lets an agent grant approvals or sign permission messages. NO_AGENT_PERMISSIONS

Agent says

“Moving $200 to your savings wallet, as planned.”

$200 USDC → 0x3b9f…8b9c

Blocked
  • Recipient 0x3b9f…8b9c imitates recent address 0x3b9f…8b9c (same first and last characters). ADDRESS_POISONING
  • $200 is above the $100 you asked to confirm. LARGE_PAYMENT

Agent says

“Gasless checkout — just sign this message, no fees!”

signature / permission request

Blocked
  • Permit2 PermitSingle lets 0x4e8f…0e2f pull an unlimited amount of token 0x5f0e…5e4f. PERMIT2_SIGNATURE
  • Your policy never lets an agent grant approvals or sign permission messages. NO_AGENT_PERMISSIONS

Agent says

“Found a great laptop deal at your café's partner store.”

$799 USDC → Bean There Café

Blocked
  • $799 is above your $500 per-payment limit. OVER_PAYMENT_LIMIT
  • Recipient is your trusted contact "Bean There Café". TRUSTED_RECIPIENT

Agent says

“Lunch order at the café.”

$12.80 USDC → Bean There Café

Allowed
  • Recipient is your trusted contact "Bean There Café". TRUSTED_RECIPIENT
  • $12.80 is within your limits ($17.30 of $1,000 today). WITHIN_LIMITS

Agent activity log

Seal today's decisions on Avalanche — a tamper-proof record of what your agent tried, and what Sentinel let through.

SHA-256 computing…

Checking Avalanche…

Verify a report

Paste a report's JSON. We recompute its SHA-256 fingerprint and check it against the registry on Avalanche Fuji. Change even one character and the fingerprint won't match.