Headline: Walrus Memory Aims to Give AI Agents Portable, Trustworthy “Agentic” Memory — Mysten Labs’ Co-Founder Explains
Mysten Labs co-founder and chief cryptographer Kostas Chalkias says the next big bottleneck for capable AI agents isn’t compute — it’s memory. In a drive to fix that, Mysten (an original contributor to Walrus) is backing Walrus Memory, a purpose-built memory layer for AI agents that promises portability, user control and coordinated, long-running context across apps and providers.
“The major misconception in AI is that compute is the only bottleneck,” Chalkias told Decrypt. “The major issue is we’re using a lot of memory as humans, and we want our LLMs to actually learn about us.” Today’s agents are cobbled together from disparate databases, vector stores and runtime state, he argues, producing brittle systems that forget or fail on complex workflows. Walrus Memory aims to solve what Chalkias calls the “real bottleneck” — agentic memory that mirrors user context and can move between sessions, apps and models.
What Walrus Memory brings
- Portability: Agents, apps and workflows can share memory independently of runtime, session or cloud provider, preventing vendor lock-in. Chalkias highlighted integrations with major models like Claude, ChatGPT and Gemini to “future-proof” workflows.
- Multi-agent coordination: Shared memory spaces let multiple agents coordinate on long-running tasks and workflows rather than operating in isolated silos.
- Cryptographic verification & programmable access: Walrus uses cryptographic tools including zk-proofs to enable contextual verification and programmable access controls over encrypted memory, letting teams protect and selectively expose data.
- Privacy and lifecycle control: Memory stored on the platform has programmable access and retention policies — “you don’t want your data to be there forever, you don’t want your data to be misused,” Chalkias said.
- Developer tooling: Plugins (OpenClaw, NemoClaw) plus Python and TypeScript SDKs make it straightforward to add portable memory to existing agent workflows.
Chalkias argued these elements are all necessary: “Just having fast compute, you don't necessarily have privacy; just having an encryption layer, you don't necessarily have a way to share your policies on whatever LLMs you want. If you just have large data, this is also not enough.” He added that, in his view, few blockchain-focused solutions currently combine all three — portability, cryptographic verification and policy-driven sharing — at once.
Early adopters and performance claims
Teams already experimenting with Walrus Memory include Allium, Conso Labs, Inflectiv, OpenGradient, Talus Labs and Tatum. Use cases range from portable agent identity systems to AI assistants that remember customer interactions across sessions.
On technical gains, Chalkias said Walrus improves the effective quality of memory provided to LLMs by addressing four layers: storage, retrieval, ranking and encryption. “In some metrics we had 60% improvements by having better ranking, better filtering and context,” he said, attributing gains to smarter classification, encryption-enabled filtering and improved relevance ranking. “We're not just a storage layer anymore.”
Why crypto and Web3 care
For blockchain and crypto builders, portable, verifiable agent memory opens new possibilities: account-linked assistant histories, auditable agent decisions, and cross-platform identity that respects user control and cryptographic guarantees. The use of zk-proofs and programmable access meshes with Web3 values around verifiability and privacy-preserving sharing.
Get started
Developers and teams can learn more or trial Walrus Memory at walrus.xyz/memory.
Sponsored: Brought to you by Walrus. Learn more about partnering with Decrypt.
Read more AI-generated news on: undefined/news
