Project Directory
Security layer for on-chain AI agents: checks every action and blocks risky trades or approvals before funds move.
BENTO is the security layer for on-chain AI agents. It sits between an agent's decision and its on-chain execution: every action is checked and analyzed first, and risky ones, such as a bad trade or an unsafe token approval, are blocked before any funds move. Sensitive operations can require the user's approval.
The rules and decision checks are encrypted with Arcium, so policies can be enforced without exposing private data. Users get a live dashboard with threat scores, policy violations and protected value, and developers integrate it through the open-source bento-sdk, a pre-execution firewall for autonomous agents. BENTO has won several hackathons, including tracks from Stellar APAC, MagicBlock, Superteam Japan and dev3pack, and is part of Superteam Japan's Colosseum hackathon cohort.