Agent Skill
Walrus Memory
An official Walrus skill for adding persistent, portable, encrypted memory to agents through TypeScript, Python, middleware, or MCP.
What is Walrus Memory?
Walrus Memory teaches an agent how to integrate MemWal rather than treating memory as a generic vector database. It explains the relationship between encrypted Walrus blobs, Sui-controlled owners and delegates, the relayer, and the operational pgvector index. The skill routes separately to TypeScript, Python, MCP, and self-hosted relayer references, and calls out failure-prone details such as asynchronous remember jobs, explicit namespaces, network-specific relayer URLs, and restore behavior. It is most useful when memory must remain portable across applications while ownership and decryption rights stay under user control.
Key capabilities
Encrypted portable memory
Stores Seal-encrypted memory blobs on Walrus while Sui contracts control owner and delegate access.
Multiple integration paths
Routes to TypeScript, Python, Vercel AI SDK middleware, OpenAI or LangChain middleware, MCP, and self-hosting guidance.
Recall and recovery model
Covers semantic recall, structured fact extraction, namespaces, and rebuilding the operational index from Walrus.
Use cases
Give an agent cross-session memory
Persist decisions or facts and recall them from several agent clients without tying the source of truth to one application.
Control the memory trust boundary
Choose managed relayer encryption, full client-side encryption, or a self-hosted relayer based on privacy requirements.
Compatibility
- A Sui wallet
- A Walrus Memory account ID and delegate key
- Node.js, Python, or an MCP-compatible client