Walrus launches MemWal SDK for verifiable agent memory

Walrus released MemWal, a software development kit that stores encrypted long-term memory records for AI agents on the Walrus decentralized data layer and provides a plugin for OpenClaw and NemoClaw. The SDK is offered by Mysten Labs’ Walrus project and is aimed at letting developers attach durable, verifiable memory to agent systems.

MemWal saves memory records in encrypted form on the Walrus network and applies programmable access controls to govern who can read or share them. Walrus’ native encryption prevents storage providers from viewing the contents, while cryptographic proofs allow recipients to verify that stored records have not been altered. The plugin for OpenClaw and NemoClaw is designed to simplify integration so teams can add verifiable long-term memory to existing agent orchestration workflows.

Abinhav Garg, group product manager at Mysten Labs, described MemWal as a way to keep memory separate from compute: “With Walrus plus MemWal, memory lives on an open, verifiable data layer, so that means it’s not tied to any one model or vendor.” He added that verifiability and tamper-resistance matter as agents handle higher-stakes tasks where correctness and auditability are required.

The project frames portability as a core feature: memory stored with MemWal can be accessed by agents running on different model providers, allowing teams to move between providers such as OpenAI and Anthropic without transferring opaque data silos. Programmable access controls let organizations set policies for sharing memory across teams or partners.

Rebecca Simmonds, managing executive at the Walrus Foundation, warned that most AI systems today rely on data pipelines that cannot be independently verified outside the organization, and she said the platform aims to provide an auditable storage layer for agent memory.

Walrus marked its one-year anniversary on March 27 and reported storing hundreds of terabytes of data on its network. The Walrus project launched soon after the Walrus Foundation raised $140 million in private funding led by Standard Crypto, with participation from Andreessen Horowitz, Electric Capital and Franklin Templeton Digital Assets.

Potential applications for verifiable agent memory cited by the company include customer support agents that retain long-term context, multiple agents coordinating using a shared memory of customer history, marketplace agents exchanging message histories as shared state, and robots that need to share operational context during extended missions such as disaster response.

MemWal is positioned as a developer-facing tool that uses the decentralized Walrus storage backbone to provide encrypted, verifiable memory records for production agent systems. A quick-start guide is available for teams that want to add MemWal memory to agents using the supported orchestration frameworks.