BetterHalf.ai

This project is a personal AI agent with decentralized, user-controlled memory. It uses Walrus for secure, permanent storage and 0g for fast memory indexing and recall. Users decide what to store and can manage their AI’s memory through tombstoning or key revocation, giving them full control over their data. This ensures context-rich personalization while keeping memory private, verifiable, and truly owned by the user. The agent can remember past interactions, files, and preferences, allowing it to provide more accurate and helpful responses over time. By separating storage from indexing, it offers efficient retrieval without compromising security. The system is designed to be modular and extensible, paving the way for future enhancements, such as integrating multiple agents or creating collaborative memory networks. Overall, it empowers users to interact with an AI that remembers intelligently while respecting their sovereignty over data.

This personal AI agent is built using a combination of decentralized storage, memory indexing, and compute orchestration. At its core, the agent uses Walrus for secure and immutable storage of user memories, ensuring that all data is decentralized and tamper-proof. To make this memory efficiently searchable and retrievable, we leverage 0g contracts for memory indexing and storage. This allows the agent to quickly query relevant memories, perform semantic searches, and maintain structured knowledge about the user’s interactions. All compute operations, including processing user input and embedding new memories, are executed via 0g compute, keeping the system decentralized and ensuring that memory operations are verifiable and trustless. The entire system is deployed on Fluence, which provides a serverless peer-to-peer compute layer, allowing agents to operate in a fully distributed environment without relying on centralized servers. We also implemented user-controlled memory management: users can tombstone or revoke keys for stored memories, effectively “deleting” them in practice even though the underlying storage is immutable. One notable hacky solution was combining Walrus immutable storage with 0g indexing: by storing encrypted memory in Walrus and indexing only the metadata and pointers in 0g, we achieved fast, verifiable memory recall while keeping user data private and decentralized. Overall, the system is modular, fully decentralized, and privacy-first, showcasing a novel approach to sovereign AI memory that can be extended in the future for multi-agent interactions or collaborative AI memory networks.