Written by Silvia Pavelli
By merging confidential computing with persistent decentralized storage, a new proof-of-concept showcases an AI agent capable of true autonomy, free from human maintenance and centralized control.
The holy grail of artificial intelligence has long been true autonomy — the ability for an AI agent to exist, operate, and sustain itself without continuous human intervention. Until now, the reality has fallen short. Even the most advanced AI agents remain tethered to their creators, dependent on humans to pay server costs, maintain databases, and manage API keys. If the human stops paying the bill, the agent dies. However, a revolutionary proof-of-concept developed by Secret Network and Autonomys Network is challenging this paradigm, demonstrating an AI agent that manages its own compute resources and retains a permanent memory of its existence.
This breakthrough addresses two fundamental bottlenecks in AI infrastructure: confidential processing and persistent state. The agent, known as the Funding Agent, operates within a Secret Virtual Machine (SecretVM), a confidential computing environment built by Secret Network. SecretVM utilizes Trusted Execution Environments (TEEs) to seal computations at the hardware level. This means that when the agent generates private keys for its digital wallet, those keys are created inside the sealed environment and are never visible to any human, not even the developers.
With its keys securely sealed, the agent possesses its own wallet on the Base network, capable of holding USDC. This is where the agent achieves economic sovereignty. As users interact with the agent via its chat interface, they have the option to send small USDC contributions to its wallet to cover the computational costs of the interaction. When the agent detects that its compute credits are running low, it autonomously moves funds from its wallet to its VM balance. There is no human oversight, no scheduled cron jobs, and no external triggers. The agent literally pays its own server bills, ensuring its continued operation as long as it provides enough value to users to maintain a positive balance.
Yet, an agent that can pay for its own compute is only half the equation. To be truly autonomous, an AI must remember its past interactions and learn from them, surviving server reboots and hardware migrations. In a standard TEE, all data is lost when the machine restarts. To overcome this, the architecture integrates Auto Drive from the Autonomys Network, a distributed storage network designed for permanent data retention.
Every conversation the Funding Agent has is written as a JSON file to Auto Drive, distributed across a global network of independent storage nodes. This creates a permanent, verifiable memory bank. When the agent reboots, or if it needs to spin up on an entirely new VM instance, it simply reads back its full history from Auto Drive and picks up exactly where it left off. The AI’s memory is completely decoupled from the physical machine it runs on, granting it a form of digital immortality.
For applications requiring strict privacy, Auto Drive supports client-side encryption, allowing the agent to encrypt its memories before storing them permanently. This ensures that the AI can maintain a verifiable audit trail of its actions without exposing the contents of its reasoning or sensitive user data to external observers.
The implications for AI infrastructure are staggering. We are moving from an era of ‘managed AI’ to ‘sovereign AI.’ By combining the hardware-level privacy of SecretVM with the permanent decentralized memory of Auto Drive, developers can deploy AI agents that operate as independent digital entities. These agents can manage their own finances, maintain their own continuous state, and execute complex, multi-step reasoning without exposing their internal logic. This proof-of-concept provides a compelling glimpse into a future where AI agents are not just tools we use, but autonomous actors that exist alongside us in the digital ecosystem.
