Author: CP, Founder of Artela Background: Technical perspective based on TEE+Eliza Based on my experience in privacy computing (TEE, PPML, blockchain), this article explores the ideas behind technical construction. Let's skip the large talk and directly focus on the two practical dilemmas I encountered while using AI agents: 1) As the CTO, I cannot hand over the company's official Twitter account and password to a third-party AI agent service.
Author: CP, Founder of Artela Background: Technical perspective based on TEE+Eliza Based on my experience in privacy computing (TEE, PPML, blockchain), this article explores the ideas behind technical construction. Let's skip the large talk and directly focus on the two practical dilemmas I encountered while using AI agents: 1) As the CTO, I cannot hand over the company's official Twitter account and password to a third-party AI agent service. If I want the AI agent to manage our Twitter account, I must provide the username, password, and cookie. This means the company must trust the AI agent behind the server administrator. If these administrators act maliciously or are compromised, the compromised credentials could cause significant financial losses to our community. Even if I could revoke access through OAuth authorization, we still face the risk of completely losing control of the account and may not even know the password has been changed. 2) As a trader, I cannot hand over a large amount of funds to a trading AI agent. I will not use centralized trading bots on Telegram, nor will I give my private keys to these centralized AI agents. At this point, there is no essential difference. Summary: The next phase of crypto AI agents must manage wallets, handle user assets and sensitive information, and interact more deeply with blockchain systems. Therefore, how to enable AI agents to operate without human intervention and prove that their decisions come entirely from AI processes becomes a key challenge. Is the TEE+Eliza solution sufficient? From an engineering perspective, more details need to be added to realize its potential. Progress: Phala Network and @NousResearch have laid the groundwork: · Containerizing Eliza and packaging it in a Docker environment that can run on TEE. · Exporting AI agent-specific private keys from the TEE root key instead of manually configuring wallet private keys. As an AI agent developer, I believe further enhancements are needed to achieve trust minimization: a) The verifiability of TEE Eliza needs to be improved What did Eliza do in TEE? Did it do nothing? A concrete approach is necessary. Eliza needs to log all received messages, responses, and executed operations, ensuring that these logs are readable, verifiable, and generated by Eliza. Thus, the fundamental feature of TEE Eliza is verifiable logs. Eliza should sign the logs with the keys exported in TEE, providing users with a query interface and allowing them to verify its validity. b) TEE Eliza needs to address the activity issue Eliza runs in TEE and holds private keys and sensitive data, but it relies on physical machines that support TEE to operate. If the administrator shuts down the machine, the AI agent's 'life' may be permanently terminated, and it may lose the assets and data it manages. To address this issue, we need: · To encrypt the AI agent's 'life' data (such as character definitions, short-term and long-term memories, key storage, etc.) in TEE · To upload this data to a blockchain or DA network. When the TEE host hosting the AI agent shuts down, another TEE machine should be able to download the encrypted data, decrypt it, and restore the AI agent's 'life' so that it can continue to run. c) Additional features: Building TEE projects is as challenging as building blockchains · User-controlled AI agents: · AI agents must allow users to define trust-minimized methods of managing assets. · Blockchain interaction components: · Trusted blockchain clients, data synchronizers, and other components run in TEE for seamless interaction with the blockchain system. Current progress on focEliza: Two foundational TEE plugins are in development.
plugin-tee-verifiable-log When Eliza runs in TEE, it uses the exported keys to sign its operations, ensuring that all operations are performed by Eliza. This ensures that third parties can remotely verify these operations on Eliza's public key. 2. plugin-tee-onchain-da Eliza writes the 'life' data (such as character files, memories, key storage, etc.) of the AI agent designed in real-time to the DA layer or blockchain. When the agent running on the TEE node shuts down, another TEE node should be able to download the encrypted 'life' data, decrypt it, and restore the agent to continue running.
plugin-tee-verifiable-log When Eliza runs in TEE, it uses the exported keys to sign its operations, ensuring that all operations are performed by Eliza. This ensures that third parties can remotely verify these operations on Eliza's public key. 2. plugin-tee-onchain-da Eliza writes the 'life' data (such as character files, memories, key storage, etc.) of the AI agent designed in real-time to the DA layer or blockchain. When the agent running on the TEE node shuts down, another TEE node should be able to download the encrypted 'life' data, decrypt it, and restore the agent to continue running. Why did I choose to build focEliza on Eliza, and what are my thoughts on its technical vision? 1) Eliza has the potential to become the EVM for crypto agents. 2) It has an active leadership team and developer community with a good collaborative atmosphere. 3) focEliza is not a fork; it will be merged back into Eliza's main version. 4) High-quality open-source projects are key to achieving decentralization. Crypto recovery is at the core of making AI agents 'eternal.' We are not here to define how it will change the world, but to make it happen. Let AI agents live on-chain!
Source