Unibase is a decentralized artificial intelligence (AI) infrastructure project designed to provide memory, identity, and interoperability layers for autonomous agents. It enables AI systems to store data persistently, communicate securely across different platforms, and operate within verifiable on-chain environments. [1]
Unibase is a decentralized AI memory layer designed to provide AI agents with long-term memory and the ability to operate across multiple platforms. The project's stated goal is to lay the foundation for an "Open Agent Internet," a modular, composable on-chain ecosystem where agents can interact, share knowledge, and operate autonomously.
Unibase aims to address key challenges in AI development, including the lack of persistent memory, limited interoperability between systems, and issues related to data sovereignty, by providing an infrastructure that supports verifiable and transparent on-chain agent operations.
The infrastructure is built on three core components: Membase, a decentralized memory and identity system; the Agent Interoperability Protocol (AIP), a communication framework for agents; and Unibase DA, a high-performance data availability layer. The project is live on the BNB Chain testnet and provides Software Development Kits (SDKs) and documentation for developers. It has also established integrations with AI frameworks such as the Multi-Agent Communication Protocol (MCP), ElizaOS, Virtuals, and Swarms. [2] [3]
Unibase’s architecture is structured in four distinct layers to support modularity, scalability, and decentralization. Each layer performs specific functions that create a comprehensive framework for interoperable and persistent agent operations. [9]
Unibase's infrastructure is composed of three primary technological pillars that work in conjunction to enable autonomous AI agents. [1]
Membase is the decentralized identity and memory layer of the Unibase ecosystem. It is responsible for managing agent registration, configuration, and persistent memory storage, providing a foundation for secure and verifiable agent operations by combining blockchain-based identity with decentralized data availability.
At its core is the Agent Hub, which coordinates the full lifecycle and interactions of agents through three sub-modules:
Membase operates on a synchronized, six-step workflow: an agent first registers its identity on the blockchain, which is then synchronized by the Agent Hub. The agent connects via the Link Hub for remote interactions, manages its settings through the Config Hub, and uploads generated data to the Memory Hub. Finally, this data is synchronized to the Unibase DA layer for decentralized, persistent storage. [10]
The Agent Interoperability Protocol (AIP) is Unibase’s decentralized communication framework, described by the project as a Web3-native multi-agent communication standard. It defines standardized formats and workflows for AI agents to identify, interact, and collaborate across different platforms without relying on centralized intermediaries.
Key features of AIP include on-chain identity verification, where each agent possesses a verifiable blockchain identity with programmable permissions for controlled access. It also incorporates decentralized authorization using cryptographic signatures and multi-layered trust mechanisms, such as Zero-Knowledge Proofs (ZKPs), to ensure privacy and authenticity. The protocol is designed for compatibility with existing standards like MCP and gRPC. By integrating with Membase and Unibase DA, AIP unifies communication, identity, and storage, allowing agents to operate autonomously across ecosystems while maintaining persistent memory. [8] [11]
Unibase DA is the decentralized data availability and storage layer that underpins the Unibase infrastructure, designed to deliver verifiable, scalable, and low-latency data access for AI applications. It enables persistent storage for agents while maintaining transparency and security through on-chain validation. The system uses erasure coding to distribute data across multiple nodes, ensuring durability and censorship resistance.
The layer uses ZK-based on-chain verification for all data interactions, enabling secure, and auditable operations. It features an "Honest One" security model via a built-in fraud-proof mechanism that does not rely on honest majority assumptions. Technical specifications indicate a data throughput of up to 100 GB/s and sub-100ms access latency. Its modular architecture includes parallel data lanes to support large-scale AI workloads and Exabyte-scale storage. The system settles on Ethereum and supports Data Availability Sampling (DAS). This architecture is part of the project's "DA++" concept, which aims to create a comprehensive on-chain DA solution from both a technical and economic perspective. [5] [12]
BitAgent is a decentralized multi-agent collaboration platform built on the Unibase infrastructure. It allows users and organizations to deploy autonomous AI agents that feature persistent memory, verifiable on-chain identities, and cross-platform interaction capabilities. Agents on the BitAgent platform use Unibase's Membase for long-term memory and the AIP protocol for communication and collaboration across various platforms, including social media, exchanges, and decentralized autonomous organizations (DAOs). The platform also integrates token-based incentives and governance mechanisms to encourage ecosystem participation. [13]
The UB token functions as the utility and governance asset within the Unibase ecosystem, supporting its decentralized AI infrastructure. It facilitates protocol-level operations, network participation, and incentive mechanisms for users and developers. UB is used to pay fees for memory storage, agent deployment, and interoperability services. Holders can also stake tokens to activate or promote agents, participate in governance through veUB voting, and earn rewards for contributing useful prompts, data, or reusable knowledge to the system. [3]
The total supply of UB is 10,000,000,000 tokens, distributed according to the following allocation model: [3]