IRSB
On-chain guardrails for AI agents
EIP-7702 delegation with five caveat enforcers — spend limits, time windows, contract and method whitelists, replay prevention — so every agent action is policy-enforced, receipted, and monitored. The same infrastructure secures intent-based DeFi solvers.
The Ecosystem
Four systems. On-chain guardrails for AI agents.
No Proof or Consequences
Protocol
On-chain contracts for receipts, bonds, disputes, and escrow. 3 verified contracts, 308 tests.
No Execution Engine
Solver
Execute intents, produce evidence, submit receipts. Pluggable job types with CLI and HTTP interfaces.
No Independent Monitoring
Watchtower
Monitor receipts, detect violations, file disputes. Rule engine with configurable violation detection.
No On-Chain Policy
Delegation (EIP-7702)
Cloud KMS signing with on-chain policy enforcement via EIP-7702 WalletDelegate and 5 caveat enforcers.
How It Works
The intent lifecycle from submission to accountability.
Submit Intent
User expresses what they want (ERC-7683 format)
Solver Executes
Bonded solver picks up and fills the intent
Receipt Posted
Cryptographic proof of execution goes on-chain
Challenge Window
1 hour for anyone to dispute with evidence
Finalize
No dispute = receipt finalizes, reputation updates
Deployed on Sepolia
3 contracts verified on Etherscan. 308 passing tests. Open source.
Where to Start
Understand the Protocol
Read the one-pager, see what changes before vs after, review the roadmap.
One-Pager →Explore the Architecture
Contracts, parameters, security model, deployed addresses, all 4 repos.
Architecture →Building an AI agent with wallet access?
We are working with agent frameworks and DeFi protocols to integrate on-chain guardrails.