Mysten Labs, the original contributor to Walrus, has launched Walrus Memory, a portable memory layer for AI agents. The platform addresses what co-founder Kostas Chalkias calls the "real bottleneck" in AI: agentic memory. Currently, developers rely on fragmented databases and runtime states, leading to unreliable systems where agents forget context across sessions. Walrus Memory enables seamless memory sharing across different apps, sessions, and AI providers without vendor lock-in. It integrates with major models like Claude, ChatGPT, and Gemini, and offers cryptographic features such as zero-knowledge proofs for privacy and programmable access control. The solution includes plugins for OpenClaw and NemoClaw, along with Python and TypeScript SDKs for easy integration. Early adopters include Allium, Conso Labs, and others building portable agent identity systems and persistent AI assistants. Chalkias emphasized that memory handling improves recall accuracy by up to 60% through better ranking, filtering, and encryption, moving beyond a simple storage layer.

Key facts

• Walrus Memory is a portable memory layer for AI agents across apps and sessions.
• Co-founder Kostas Chalkias says agentic memory is the real bottleneck in AI.
• Integrates with Claude, ChatGPT, and Gemini, avoiding vendor lock-in.
• Uses zk-proofs and programmable access control for privacy and security.
• Improves recall accuracy by up to 60% through better ranking and filtering.

KeyAudit data perspective