Open Console →
● Human always in control

Autonomy you can actually trust

"Autonomous" should never mean "unaccountable". Every safety property here is enforced by code on a public blockchain — not by a promise. Here's exactly how you stay in control.

The one idea that makes it safe

The agent never holds the key that owns everything

🗝️ Master key — you

Stays in your wallet. It's the only key that can change the rules, move everything, or take the agent back. The runtime never sees it.

🎟️ Session key — the agent

A temporary, scoped key the agent's runtime uses to act. It expires, it's spend-limited, and you can revoke it any time — without moving a single coin.

Five layers between an agent and your money

💳
Hard spend limits

Per-transaction, per-day and per-month caps — for native RAMA and for each token — enforced on-chain. The agent cannot exceed them, full stop.

Allow-lists & capabilities

A session key can only reach the exact contracts, recipients and actions you approve. It cannot approve tokens or hit arbitrary routers.

Human approval inbox

Anything above a threshold you choose pauses and waits for your explicit "yes" — bound to the exact call, so an approval can't be reused for something else.

⏸️
Instant pause & revoke

Freeze the agent or cancel its session key in one click, from anywhere. No waiting, no counterparty.

🔎
Everything on the record

Every action is written to a public blockchain you can inspect and verify. There are no hidden, off-ledger moves.

🏛️
24-hour upgrade timelock

Even the system's own code can only change after a public 24-hour delay via an on-chain timelock — no instant or secret upgrades.

Governance

Who can change the system — and how slowly

The contracts are upgradeable proxies (so bugs can be fixed without new addresses), but that power is deliberately constrained: a 2-of-3 operations multisig can only propose a change to an on-chain timelock, which then enforces a 24-hour public window before anyone can execute it. Bad or unexpected upgrades can be seen — and cancelled — before they land.

🔑 3 signer keysheld by operators 🖇️ 2-of-3 multisigproposes ⏳ 24h timelockpublic delay 📜 Contractsupgrade

Straight talk

What we're not claiming

⚠️

Guarded beta on mainnet. The contracts are live and hardened, and an independent internal review has been done — but an external third-party audit is still pending. Use small test allocations only, keep autonomous actions tightly scoped, and always keep a human approval path for anything sensitive. Blockchain actions are irreversible; never let an agent handle funds you can't afford to lose.

Start small, stay in control

Give one agent a tiny budget and a single job. Watch every move. Grow only when you're comfortable.

Found a vulnerability? Please do not open a public issue — email [email protected] or see our responsible-disclosure policy.