Let your AI agent spend real money. Keep the leash.
Your agent works inside limits a hijacked agent cannot exceed. Watch it reason live. Step in when it matters. Snap the leash when it doesn't.
Refused. Over the per-transfer cap.
The contract held. You didn't have to be awake.
Revoked in one move.
The record stays: hash-chained, yours alone.
Simulated. The real leash lives in the cockpit.
The limits, in plain numbers
These are the live limits on the leash account in the receipts below.
This already went wrong for other people.
In January 2026, ungoverned trading agents at Step Finance moved 27 to 30 million dollars. A prompt injection drained roughly 150 to 200 thousand dollars from Grok and Bankr users after a guardrail was silently rewritten. The lesson is simple.
Instructions don't survive a clever prompt. Contracts do.
So LEASH puts the boundary where prompts can't reach it: enforced by a contract on 0G Chain, not by a system prompt the agent can be talked out of.
How the leash works
Bind
Create a leash account for your agent: a per-transfer cap, a spending window, an allowlist, and an expiry. The limits are enforced on 0G Chain.
Watch
See its reasoning live as it's served by 0G Compute, with attested inference where you choose the TEE providers.
Contain
A fully hijacked agent cannot exceed the caps. Over the limit, the contract refuses. Step in before it spends, or don't. The boundary holds either way.
Prove
Every decision is hash-chained, encrypted to you alone, and anchored on 0G Storage. Revoke in one move. The record stays.
Most days it's one small question.
A rare boundary decision arrives with Approve and Deny inline. A daily digest tells you what was spent and what was refused. Nothing else asks for your attention.
The same card lands on your phone, through the LEASH Telegram bot.
The receipts
- AgentRegistry0xA74d…42C3sourcify-verified
- LeashAccountFactory0x5100…3c52sourcify-verified
- A coordinated agent act0xf3e6…a34fconfirmed transaction
Live on 0G testnet. No mainnet funds are at stake yet.