Walrus launches Memory SDK to give AI agents portable, encrypted

Walrus, the decentralized storage protocol built on the Sui blockchain, has released the MemWal SDK—a developer toolkit that provides AI agents with persistent, encrypted memory and semantic search on Walrus’s network [1]. The beta‑stage SDK aims to solve the “agentic memory” bottleneck by making context portable, verifiable and shareable across providers.

Key takeaways

• MemWal stores encrypted memories on Walrus and adds semantic‑search retrieval, letting agents query by meaning rather than exact keywords [1].
• The platform integrates with major models (Claude, ChatGPT, Gemini) and offers plugins for OpenClaw and NemoClaw, plus Python and TypeScript SDKs [2][3].
• Memory is encrypted by default and governed by programmable access controls, allowing users to decide who can read or write the data [1][3].
• Walrus leverages the Sui blockchain for ownership and access control, ensuring that memory remains user‑owned rather than tied to any single AI provider [1].
• Early adopters such as Allium, Conso Labs, Inflectiv, OpenGradient, Talus Labs and Tatum are building applications that use the memory layer for persistent agent identity and cross‑session assistance [2][3].

Decentralized memory for AI agents

The MemWal SDK layers semantic search on top of encrypted storage, so agents can retrieve relevant context based on meaning instead of exact text matches [1]. Walrus’s infrastructure, built on Sui, handles ownership and access control, giving users the ability to grant or revoke permissions for reading, writing, or sharing an agent’s memories [1]. The SDK ships with integrations for the Vercel AI SDK and includes plugins for the OpenClaw and NemoClaw frameworks, with documentation and a GitHub feedback channel to support developers during the beta phase [1].

Mystic Labs co‑founder Kostas Chalkias describes the offering as a response to the “real bottleneck” of agentic memory, noting that current AI workflows rely on ad‑hoc combinations of databases, vector stores and runtime state, which often fail in long‑running or multi‑agent scenarios [2]. Walrus Memory (the product name used in public communications) adds cryptographic tools such as zk‑proofs for contextual verification and programmable access control, aiming to provide verifiability, availability, portability and shareability—the four pillars highlighted by the Walrus team [1][2][3].

Why it matters

By moving AI memory to a decentralized, verifiable storage layer, Walrus seeks to eliminate the “Goldfish Problem” where agents lose context after a session ends or a runtime restarts [4]. Persistent, portable memory enables agents to coordinate across different models and providers, reducing the need to rebuild context when switching between platforms like OpenAI, Anthropic or Gemini [3][4]. As developers adopt the SDK, the ecosystem could see more robust multi‑agent workflows, improved recall accuracy, and greater user control over data privacy and longevity. Walrus has made the SDK available now, with free access for developers at walrus.xyz/memory, positioning the platform as a foundational piece of emerging .