Policy checks before agents act.
Define deterministic rules for AI agents. Evaluate requests, log decisions, and verify everything on-chain. No AI in the loop, no ambiguity.
Policies
Select a policy to evaluate against, or create your own.
Rules for Safe default
Risk score must stay at or below 4
Hackathon deployments should prefer visibly conservative automation.
Tool cannot be shell/exec
Direct shell execution is the most obvious breakout vector.
Target cannot reference a production domain
Keep early agent actions away from production systems.
Action type must be read, classify, or summarize
Only deterministic, low-impact actions are allowed in this profile.
Evaluate
Test an agent action request against the selected policy.
What the agent wants to do
Which tool the agent uses
0 = no risk, 10 = critical
System or resource being accessed
Audit log
No audit entries yet.
Submit an action request to create the first entry.
Verify a decision
Re-evaluate a request against a policy, hash it, and check if the decision exists on-chain.
On-Chain Tools
Register agent identity, form bilateral agreements, and manage spending vaults.