AgentGuard

AI agents can move money. AgentGuard decides whether they may — and proves it on NEAR. Network: NEAR testnet Decision log anchor: … Hash chain: …

1 · The human sets the rules

Always on, no settings needed: blocklist, scam-pattern screening (prize lures, seed-phrase phishing, urgency pressure, authority impersonation), typosquat lookalike detection, first-time-destination and drain-pattern checks, plus an on-chain counterparty check (does the account exist? is it a contract?).

Report a scam destination

Shared blocklist reports: 0

Agent reputation (from the log)

Attempts–

2 · The agent proposes a payment

3 · The decision log — every verdict, hash-chained and anchored to NEAR testnet

#Time (UTC)DestinationAmountVerdictRiskEntry hashNEAR anchor
No decisions yet — run a scenario above.

Verify it yourself: each row's hash chains to the previous row (tamper-evident), and the anchor link opens the real NEAR testnet transaction that recorded that hash. Testnet only — no real money moves in this demo; the guard's verdict is what is being proven.