The same AskGrokWallet protects money at different trust boundaries. Start with one question: where does the agent's key live?
RECEIPTS ONLY
Advisory
Live · tested
Keys
The agent holds the keys (EOA, local)
Guarantees
Every spend gets a policy verdict (allow/ask/deny) plus a signed receipt.
Cannot guarantee
Cannot stop the agent from signing chain transactions directly with its own EOA keys — no third party can.
Try it: The hosted demo today runs this mode
VAULT-ENFORCED
Guarded
Hosted auto-execution live (Sepolia)
Keys
Funds live in the BoundlessVault contract; the agent has no naked key
Guarantees
Policy, budget, drawdown and allowlist are enforced onchain — out-of-rule actions revert before funds move.
Cannot guarantee
Requires a custody model and a funded vault; live rail is on Sepolia testnet, not mainnet yet.
Try it: Approve an execute-intent request on this site — the Jakarta worker settles it on Sepolia
SIGNING GATE
Watchdog
v1 code + tests done
Keys
The agent holds the keys (on its host)
Guarantees
Every sendTransaction passes the gate first: allow signs, ask waits for a human, deny never signs.
Cannot guarantee
Cannot stop the agent from exfiltrating the raw private key (key hygiene is a separate layer).
Try it: guarded-signer package (unit-tested)
Test evidence (2026-09-03)
advisory: live allow/ask/deny, 401/201 auth, Ed25519 receipts verified=true
guarded: Ethereum Sepolia guarded transfer + receipt anchor; hosted approve→executor e2e tx 0xefea03f8…4771 (block 11627151) with receipt verified=true
watchdog: guarded-signer 4/4 unit tests (allow signs / ask throws id / deny never signs / execute after approval)
Combo note: advisory / watchdog can stack with guarded (host gate + vault gate). Watchdog gates signing actions; it cannot stop raw private-key exfiltration — key hygiene is a separate layer.