Mnemo

Lack of interoperability remains a fundamental limitation across both blockchain networks and AI systems. Today, each chain operates in isolation, and similarly, each AI model exists within its own closed environment. Conversations, preferences, and learned context are not portable—what one model learns about a user cannot be transferred to another. As a result, users repeatedly rebuild context, and developers repeatedly rebuild memory systems. Even within a single platform, memory is often inconsistent or session-bound, making AI agents effectively “amnesiac.”

Mnemo addresses this by introducing a persistent, portable memory layer for AI agents. Built on 0G Protocol, Mnemo enables users to own and control their long-term context as a verifiable, decentralized resource. Memory is stored as encrypted chunks on decentralized storage, processed through semantic retrieval using distributed compute, and governed by on-chain access control. This ensures that memory is not only persistent but also secure, permissioned, and interoperable across applications.

With Mnemo, users can define and retain important context such as preferences, behavioral patterns, rules, and frequently used actions (e.g., transaction recipients or workflows). Any AI agent or application that is explicitly authorized by the user can access this memory layer, interpret it, and act on it without requiring retraining or manual reconfiguration. This transforms memory into a shared primitive: write once, use everywhere.

By decoupling memory from individual applications and anchoring it in decentralized infrastructure, Mnemo enables a new class of intelligent, context-aware agents that persist across platforms. It removes the friction of starting over and lays the foundation for a true agent economy where intelligence compounds over time and remains fully owned by the user.

Mnemo is built as a full-stack application combining a modern web interface, on-chain access control, decentralized storage, and AI-powered agents. The frontend is developed using Next.js, which handles the user interface for interacting with agents, viewing stored memory, and managing permissions. Reown's wagmi library handles the interactions betweenthe frontend and the smart contract.

On-chain logic is written in Solidity and deployed on the 0G Galileo Testnet. The smart contracts manage agent registration, as well as access control allowing users to grant or revoke specific agents’ ability to read or write to their memory vault. This ensures that ownership and permissions are enforced transparently on-chain.

For storage, Mnemo uses the 0G TypeScript SDK to upload and retrieve memory data. User context (such as preferences, instructions, or past interactions) is stored as structured data and linked to the user’s identity - their wallet address. Instead of keeping memory inside a single application, it is externalized and made accessible across different interfaces.

On the AI side, the OpenAI SDK is used to power conversational agents. The agents are designed to interpret user input, decide when to store relevant context, and retrieve previously saved information when needed. Memory recall is currently implemented through direct lookups and simple matching rather than semantic search, which keeps the system lightweight while still demonstrating cross-session persistence.

A notable design choice was separating memory from the agent itself. Instead of tying context to a single app session, all memory is stored independently and referenced when needed, making it portable across different agent interfaces. This modular structure allows Mnemo to function as a shared memory layer rather than just another AI chat application.