In a major development at the intersection of blockchain and artificial intelligence, Sui-based storage protocol Walrus has officially launched MemWal, a memory layer and SDK product designed specifically for AI agents. This launch, reported by Decrypt, marks a crucial step toward creating a decentralized, verifiable memory infrastructure for autonomous AI systems.
Walrus MemWal: A new memory paradigm for AI agents
MemWal provides AI agents with verifiability, availability, portability, and shareability for their memory. Abinhav Garg, product manager at Mysten Labs – the developer of Sui and Walrus – explained that using Walrus and MemWal together stores memory on an open and verifiable data layer. This eliminates the dependency on a single AI model or provider.
This approach allows users to freely switch between AI models such as ChatGPT and Claude. It also enables new applications that can remember user-specific signals across platforms and sessions.
Key features of MemWal
- Verifiability: All memory stored on Walrus is cryptographically verifiable, ensuring data integrity and provenance.
- Availability: Data remains accessible as long as the Walrus network is active, without any point of failure.
- Portability: Users can move their AI agent’s memory between different models and applications without data loss.
- Divisibility: Memory can be selectively shared with other agents or applications, enabling collaborative AI workflows.
How Walrus and MemWal work together
Launched on Sui’s mainnet in late 2024, Walrus offers decentralized blob storage optimized for large data objects. MemWal builds on this foundation by adding a structured memory layer specifically for AI agents. The SDK provides developers with tools to read, write, and manage agent memory in a decentralized manner.
This architecture addresses a critical challenge in AI development: the lack of persistent, portable memory across models and platforms. Currently, most AI agents operate in isolated environments, losing context when switching between models or applications.
Technical Architecture
MemWal uses Walrus’ blob storage to store memory objects. Each memory object contains metadata such as timestamps, ownership, and access control. The SDK handles encryption, indexing, and retrieval, allowing developers to easily integrate persistent memory into their AI agents.
The system supports multiple memory types, including conversation history, user preferences, task states, and learned behavior. Developers can define custom memory schemes to suit their specific use cases.
Impact on the portability of AI models
One of the key implications of MemWal is its potential to break down AI walled gardens. Currently, users are often tied to a single AI provider because their data, context, and preferences are stored within that provider’s ecosystem.
MemWal allows users to maintain consistent memory across different AI models. For example, a user can start a conversation with ChatGPT and then continue seamlessly with Claude, with both models accessing the same memory storage. This interoperability could accelerate AI adoption by reducing switching costs.
Real-world use cases
- Personal AI Assistants: Ensure consistent user preferences and conversation history across AI platforms.
- Enterprise AI agents: Share context and learned behavior with multiple agents working on the same project.
- Gaming AI: Ensure NPCs can remember player interactions across gaming sessions and platforms.
- Healthcare AI: Maintain patient context through various diagnostic and treatment planning tools.
Market context and timeline
The launch of MemWal comes as the AI industry struggles with the limitations of current memory architectures. Major AI providers like OpenAI, Anthropic, and Google have all announced efforts to improve context windows and memory capabilities, but these remain proprietary and platform specific.
Walrus’ decentralized approach offers an alternative that prioritizes user control and data portability. The project has gained significant popularity since the mainnet launch, with more than 1,000 developers already building on the platform.
Expert perspectives
Abinhav Garg highlighted the philosophical shift behind MemWal: “We believe that AI memory should be owned by users and not locked into a single provider. MemWal gives users the freedom to choose the best AI for each task, without losing context.”
Industry analysts have noted that this approach aligns with increasing regulatory pressure for data portability and interoperability in AI systems. For example, the European Union’s AI law includes provisions for user data rights that could benefit from decentralized memory solutions.
Technical considerations and challenges
While MemWal offers significant benefits, it also has challenges. Decentralized storage introduces latency compared to centralized solutions, which can impact real-time AI interactions. The Mysten Labs team has implemented caching and optimization strategies to mitigate this.
Another consideration is cost. Walrus uses a storage market where users pay for data persistence. While costs are competitive with centralized alternatives, they can become significant for applications with large memory requirements.
Security and privacy
MemWal includes encryption at rest and in transit, with users controlling access via cryptographic keys. This ensures that even though the memory is stored on a public network, only authorized parties can access it. The system also supports selective disclosure, allowing users to share specific memory segments without exposing their entire history.
Comparison with existing solutions
Future roadmap
Mysten Labs has outlined an ambitious roadmap for MemWal. Near-term plans include integration with major AI frameworks such as LangChain and LlamaIndex. The team is also working on performance optimizations to reduce latency to levels competitive with centralized solutions.
In the longer term, the project aims to become the standard memory layer for decentralized AI agents. This includes support for multi-agent memory sharing, memory state versioning, and integration with decentralized identity systems.
Community and ecosystem
The Walrus community has responded positively to the launch of MemWal. Several projects have already announced plans to integrate the SDK, including decentralized AI marketplaces and personal assistant applications. The open-source nature of the project encourages community contributions and third-party development.
Conclusion
The launch of Walrus MemWal represents a significant advancement in the quest for decentralized, portable AI agent memory. By providing verifiability, availability, portability, and shareability, MemWal addresses critical limitations in current AI architectures. As the AI industry continues to evolve, solutions like MemWal that prioritize user control and data portability will become increasingly important. Developers and users alike should keep a close eye on this space, as MemWal has the potential to reshape how we interact with AI agents across platforms and providers.
Frequently asked questions
Question 1: What is Walrus MemWal?
MemWal is a memory layer and SDK product launched by Walrus, a Sui-based storage protocol. It provides verifiable, portable, and shareable memory for AI agents, allowing them to maintain context across different models and applications.
Question 2: How does MemWal improve the functionality of AI agents?
MemWal allows AI agents to store and retrieve memory in a decentralized manner, eliminating dependency on a single AI provider. This allows users to switch between models like ChatGPT and Claude without losing context.
Question 3: Is MemWal compatible with existing AI frameworks?
Yes, the SDK is designed to integrate with popular AI frameworks. The team is actively working on integrations with LangChain, LlamaIndex and other key tools.
Question 4: How does MemWal ensure data privacy?
MemWal uses encryption at rest and in transit, with user-managed access keys. It supports selective disclosure, allowing users to share specific memory segments without exposing their entire history.
Question 5: What are the costs associated with using MemWal?
The costs are based on the Walrus storage market, where users pay for data persistence. While competitive with centralized alternatives, costs can vary depending on the amount of memory stored and the duration of storage.
Question 6: Can MemWal be used for enterprise applications?
Absolute. MemWal is designed for both individual and enterprise use cases, including multi-agent collaboration, business AI assistants, and complex workflow automation.