Introducing Sigilla - Composable Permissions for Autonomous Agents on SUI.

Web3 AI agents are stuck at the approve wall: you either sign every action yourself, or hand the agent your keys and hope. Neither works when real money's on the line. Sigilla is the missing safety layer for autonomous on-chain agents: you compose AI agents on a canvas, and the Sui VM itself enforces what each agent is allowed to do so even a compromised agent or backend can't drain you, and one click revokes it. The core premise: permissions belong in the blockchain, not in the servers. Each agent acts through a Permit which is a Move policy object that gates every action (coin withdrawals, Deepbook swaps, order-book trades). The agent only holds a revocable permission cap, the Permit itself has the final say on every move. Sui-native by design. This isn't a port of an EVM idea. It's only possible because of Sui's object model + PTBs: policy lives in an owned/shared object. Live on Sui testnet: Move package, typed SDK, Next.js canvas UI, event indexer, autonomous agent runtime, and Deepbook v3 adapters. Try it now at sigilla.xyz