Throughout the last quarter of 2024, AI Agents were the talk of the town, dominating on-chain mindshare and market share. AI agents are intelligent, autonomous virtual assistants designed to relieve us of tasks that would otherwise require human effort. To read more about AI Agents, refer to my previous piece that serves as a primer to this intriguing new crypto vertical.

Since then, the agentic vertical has evolved leaps and bounds. Gone are “sentient yappers” and the automated reply guys. The flavour of the market has since rotated to agentic projects with utility or have a defining innovative feature that serves as a moat. 

The majority of AI Agents from the earlier part of Q4 ‘24 have now been forgotten. There are only a number of AI Agents from that time that have been able to establish a moat, such as Zerebro, who made music and art its niche, or aixbt, who brands itself as the KOL who gives alfa 24/7. An up and coming segment of the agentic market are AI-powered applications. An example is the DeFAI segment, which leverages the automation and sentience of AI Agents to smoothen users’ DeFi experience across various dApps (decentralised applications) and functionalities. Another example are the projects that are trying to integrate AI Agents into games, giving existing games a new flair. Lastly, AI Infrastructure such as launchpads (e.g. Virtuals, Top Hat) and development frameworks (e.g. ai16z, Zerebro) have taken the spotlight, seeing the most capital inflow as of late.

Where does TEE fit in all this?
An overlooked layer of the agentic tech stack is the integration of Trusted Execution Environments (TEEs). A TEE is a specialised hardware designed to create isolated environments for running sensitive computations securely. It ensures data confidentiality and protection even in untrusted or potentially compromised systems. TEEs act as a safeguard, isolating critical applications (such as private keys, AI models, or sensitive credentials) from less secure or unrelated applications. Communication between the secure and normal environments is tightly controlled and allowed only with specific privileges.

In the context of AI agents, the agent and its private key can operate entirely within the secure environment of a TEE. This setup prevents unauthorized access or abuse of the agent or its credentials. It ensures that even if someone tries to compromise the broader system, they would need special privileges to interact with or exploit the secure AI agent. This capability is particularly valuable in decentralized AI systems, where trust, privacy, and security are essential for handling proprietary or sensitive information.

Phala Network
Phala Network is an example of projects building in the TEE vertical. It is a protocol that revolves around verifiable compute, made possible by TEEs. Phala’s TEE technology combines software and hardware-based TEEs, enhancing security and privacy for decentralized AI (DeAI). Additionally, Phala's Key Management System (KMS) enhances security by continuously rotating management keys, reducing the risk of prolonged data exposure and providing an added layer of protection.

More relevant to AI Agents, a key development for the AI Agent vertical was Phala’s TEE integration into ai16z’s Eliza, the leading open-source AI Agent framework on Solana at present. Eliza allows for the deployment of AI agents across social media platforms such as Discord, Twitter, and Telegram. It offers a RAG memory system, media analysis, document processing, and more. By integrating Phala’s TEE, these capabilities benefit from cryptographic verification and improved privacy. Phala’s framework supports modularity and custom plugins, allowing developers to scale seamlessly and incorporate new integrations with ease.

On the roadmap is Phala 2.0, which introduces a powerful GPU TEE network, an upgrade from its existing CPU-based infrastructure. By integrating NVIDIA Confidential Computing (TDX), Phala ensures sensitive GPU computations remain isolated and secure, even on compromised hosts. Additionally, Phala is launching a Layer 2 rollup on Ethereum to provide fast, low-cost, and secure transactions while combining scalable ZK proofs with TEE infrastructure. Through Phala Cloud, developers gain seamless access to GPU and CPU TEEs, enabling private, AI-ready application deployment with real-time monitoring and trust verification.

Phala has seen a good amount of adoption across its product suite. Service providers such as Hyperbolic Labs, io.net, NEAR Protocol and more have adopted Phala’s TEE GPUs to secure LLM model training and data. On the other hand, 0G Labs, Ritual and Redpill have been using Phala’s TEE CPUs to safeguard LLM API calls.

The future of TEE agentic integrations
As the AI Crypto vertical continues to grow, it is likely that we see more use-cases and synergies between agentic developments and the current crypto tech stack. One such possible development in the near future is Eigenlayer’s potential involvement in furthering verifiable and secure computation. 

Most of us might know Eigenlayer as the premier infrastructural layer that allows for the restaking of Ethereum and its Liquid Staking equivalents for projects to tap on Ethereum’s security. However, recently, the Eigenlayer team seems to have taken an interest in AI Agents. We think that there is an opportunity for Eigenlayer to contribute to the agentic tech stack by providing AI Agents with a decentralised compute stack secured by Ethereum, via Eigenlayer’s Actively Validated Services (AVSs). 

This is just the tip of the iceberg. We believe that there will be a plethora of existing projects that will bring useful integrations to AI Agents, embracing the collaborative and open-sourced nature of the crypto community

Thank you for reading!