Tee Agent is an Arbitrum-native trust passport for AI agents. It lets agents be minted as on-chain entities with ERC-8004 identity and reputation, ERC-7857 encrypted private data, and Phala Intel TDX oracle execution verified through Automata DCAP.
AI agents are starting to own wallets, call tools, and make decisions for users, but trust is still missing. Who owns an agent? What services does it expose? Was its output produced by the registered oracle? Has the result been validated? Is feedback genuine or just spam?
Tee Agent solves this by extending ERC-8004 with a deployed validation registry, validation-backed feedback, and encrypted ERC-7857 agent data. Instead of using a TEE only as a validator, Tee Agent uses the Phala TDX oracle as the agent runtime: it decrypts private skills/files inside the enclave, runs the agent, validates outputs, handles transfer re-encryption, and produces proofs that can be checked on-chain.
The demo flow mints an AI agent on Arbitrum Sepolia, runs it through a Phala CVM oracle, requests validation, submits feedback linked to that validation, and verifies the feedback through the dashboard or MCP. AI clients can also use the MCP server to discover agents, prepare mint/validation/feedback transactions, read reputation, and verify feedback without handling ABIs directly.