Most AI agents have a hidden dependency. Behind every chatbot, assistant or autonomous system sits a human operator paying server costs, maintaining databases, managing keys and ensuring the infrastructure remains operational. Remove that support and the agent disappears.
A new proof-of-concept developed by Secret Network and Autonomys explores a different model. The demonstration features an AI agent capable of managing its own operating costs while maintaining a persistent memory across restarts, creating a system that moves a step closer to genuine autonomy.
Called the Funding Agent, the experiment combines Secret Network’s confidential computing environment, SecretVM, with Autonomys’ distributed storage platform, Auto Drive. Together they provide two of the most significant ingredients required for long-lived AI systems: private computation and permanent memory.
The agent itself is relatively simple. Running inside a SecretVM, it operates with its own wallet on the Base network, holding contributions sent by users in USDC. When its computing resources begin to run low, the agent automatically transfers funds from its wallet to cover the cost of continuing operations. No administrator schedules the payment and no external script initiates the transaction. As long as sufficient funds remain in its wallet, the agent can continue paying for the resources it requires.
While that may sound straightforward, it addresses one of the central challenges facing autonomous AI systems. Most agents today rely on infrastructure controlled by humans. Even sophisticated agents capable of making decisions and performing tasks remain dependent on someone else paying the bills.
The demonstration explores what happens when an agent gains direct control over its own operational budget.
At the heart of the system is SecretVM, Secret Network’s confidential computing environment. Unlike conventional cloud infrastructure, SecretVM runs inside a trusted execution environment where computations are sealed at the hardware level. Data remains encrypted both in transit and at rest, while the processor itself enforces isolation between the virtual machine and the outside world.
For AI systems, this has an important consequence. Private keys generated within the SecretVM never become visible to any human operator. They are created, stored and used entirely within the protected environment.
The second component of the architecture addresses a different problem: memory.
AI agents frequently lose context when infrastructure changes, servers restart or deployments are replaced. In this proof-of-concept, every conversation is written as a JSON file to Auto Drive, the distributed storage layer developed by Autonomys. Rather than storing memory on a single machine, the information is distributed across a network of independent storage nodes.
When the agent restarts or a new virtual machine instance is created, it can retrieve its previous records and continue from where it left off. Its memory survives independently of the hardware on which it is running.
The combination creates an interesting architecture. SecretVM provides confidential computation, ensuring the agent’s reasoning and private keys remain protected. Auto Drive provides permanent storage, preserving conversation history and operational records beyond the lifespan of any individual machine.
Separately, each technology addresses part of the problem. Together they create a framework in which an agent can operate privately while maintaining continuity over time.
The proof-of-concept has been deliberately designed for public inspection. The agent’s conversation history and stored memory files are visible, allowing developers to verify the claims being made. However, the architecture also supports encrypted storage for production environments where privacy is required. Data can be encrypted before being uploaded to Auto Drive, preserving both confidentiality and verifiability without exposing the underlying content.
The implications extend beyond a single demonstration. Long-term autonomous agents have been discussed extensively within both artificial intelligence and blockchain communities, yet practical implementations remain rare. Persistent memory, secure identity management and financial self-sufficiency are frequently cited as prerequisites for agents capable of operating independently over extended periods.
While the Funding Agent remains an experimental project, it provides a glimpse of how such systems might function in practice.
Rather than existing as temporary software processes tied to a particular server or deployment, future agents may increasingly resemble digital entities with their own resources, histories and operational continuity. They would be capable of preserving memory across infrastructure changes, managing their own budgets and continuing to operate without requiring constant human intervention.
For now, the Funding Agent is a modest demonstration. Yet it raises a larger question about the future of autonomous systems. If an AI agent can securely hold funds, pay for its own computing resources and retain its memory indefinitely, the distinction between software application and independent digital actor begins to blur.
