Headline: Mysten Labs’ Walrus Memory aims to give AI agents real, portable “memories” — with crypto-grade privacy and cross-model support
Mysten Labs co-founder and chief cryptographer Kostas Chalkias says the missing piece for reliable AI agents isn’t raw compute — it’s memory. Today Mysten, an original contributor to the Walrus project, is unveiling Walrus Memory: a dedicated memory layer for agentic AI designed around portability, user control and agent coordination.
Why memory matters
- Current agent stacks stitch together ad-hoc databases, vector stores and runtime state. The result: brittle systems that struggle with multi-step workflows and routinely “forget” context.
- Chalkias argues the misconception in AI is that compute is the bottleneck. In practice, agents need to “actually learn about us” — i.e., keep and manage usable memories across sessions and apps.
What Walrus Memory does
- Portable memories: agents, apps and workflows can share context without being locked to one runtime, session or model provider.
- Shared memory spaces: multiple agents can coordinate across long-running workflows by accessing common context.
- Cryptographic verification and programmable access: primitives such as zk-proofs are used for contextual verification and for enforcing access policies over encrypted memories.
- Cross-model integrations: Walrus Memory works with leading LLM platforms including Claude, ChatGPT and Gemini so users aren’t tied to a single provider.
- Programmable controls: data stored on Walrus supports fine-grained access policies and lifecycle rules (so memories aren’t retained or misused indefinitely).
Chalkias: three-pronged solution
Chalkias says solving agentic memory requires more than speed, encryption, or raw data alone — you need all three: privacy, policy-sharing across models, and smart data handling. “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,” he said. “If you just have large data, this is also not enough.”
Developer tooling and early adopters
- Plugins and SDKs: OpenClaw and NemoClaw plugins plus Python and TypeScript SDKs let developers add portable memory to existing agent workflows.
- Early partners: teams including Allium, Conso Labs, Inflectiv, OpenGradient, Talus Labs and Tatum are using Walrus Memory to build things like portable agent identity systems and assistants that remember customer interactions across sessions.
Performance lift and scope
Walrus Memory targets four core areas that affect LLM memory quality: storage, retrieval, ranking and encryption. Chalkias says improved ranking, filtering and encrypted data handling produced as much as a 60% improvement on some metrics, arguing the product is “not just a storage layer anymore.”
Why this matters for crypto and web3
The combination of portable memory, programmable access control and cryptographic verification aligns with web3 principles: user control of data, verifiable context, and interoperability across providers. Chalkias notes he doesn’t see any blockchain-focused solution currently addressing all three elements together — a gap Walrus Memory aims to fill.
Learn more or get started at walrus.xyz/memory. Brought to you by Walrus.
Read more AI-generated news on: undefined/news
