Blind Agent Framework

Midnight Blind Agent: Securely automate corporate governance with Gemini AI and ZK-privacy. Manage assets autonomously on the Midnight network without exposing sensitive business data.

Comment

Inspiration

In standard blockchain architectures, business logic and data are public. For enterprises, this is a barrier; you cannot expose treasury balances or supplier prices to competitors. We were inspired by the Midnight Network to build a system where AI can manage a company "blindly"—executing valid logic while keeping the underlying data encrypted.

What it does

The framework creates an autonomous "Enclave." Procurement, Treasury, and Compliance agents (powered by Gemini) analyze private variables to make decisions. Instead of publishing sensitive data, they submit Zero-Knowledge (ZK) Proofs to the public ledger to attest that their actions were authorized and valid.

How we built it

We built a custom Midnight Protocol Simulator in TypeScript to model the dual-state (private vs. public) architecture. We used Google Gemini as the agent's "Mind," processing private telemetry to output ZK-attested transactions. The UI uses Framer Motion and Tailwind to visualize the separation between the secure enclave and the public network.

Challenges we ran into

The biggest challenge was architecting the "Selective Disclosure" mechanism. We had to ensure that the Viewing Keys could decrypt specific private variables for auditors without leaking the entire operational history of the autonomous agents.

Accomplishments that we're proud of

We successfully visualized the flow of Private State vs. Public Proofs. Seeing an AI agent reorder inventory based on an encrypted threshold and seeing only a "CONFIRMED" proof hit the public table was a "Eureka" moment.

What we learned

We learned that the future of enterprise AI isn't just about intelligence, but attested privacy. Midnight’s concept of "witnesses" and "reveals" perfectly complements autonomous agent workflows.

What's next for Blind Agent Framework

We plan to transition from our TypeScript simulator to the actual Midnight .compact language, enabling real-world shielding of AI-driven business transactions on the testnet.

Built With

Share this project:

Updates