Walrus Memory

Walrus Memory is a product from Walrus — a verifiable data platform created by Mysten Labs, the team behind the Sui blockchain — designed to give AI agents portable, persistent, and cryptographically verifiable memory. It sits within the broader Walrus ecosystem, which uses decentralized storage nodes and onchain proofs to make data always available and independently verifiable.

What It Is

Walrus Memory is a specialized memory layer for AI agents, addressing the challenge of agents losing context between sessions or across different environments. By building on Walrus's decentralized blob store — which uses Sui for coordination and governance — agent memory becomes portable (not locked to a single runtime), verifiable (every stored blob gets a verifiable ID and tracked history), and resilient (data remains available even if a significant portion of storage nodes fail). The core Walrus platform is open source under the Apache License 2.0, with the repository hosted at MystenLabs/walrus on GitHub.

How It Fits in the Walrus Stack

Walrus Memory is one of the named products within the Walrus platform, alongside the broader infrastructure for AI, Data Markets, and DeFi use cases. The underlying Walrus platform provides:

• Decentralized storage nodes — data is spread across nodes, resilient to failure even if two-thirds of nodes go offline
• Fast reads and writes — a global network reconstructs and delivers data in milliseconds; the site notes approximately 800ms downloads for small files under 1MB
• Publicly verifiable proofs — onchain proofs let anyone verify data independently, at any time
• Programmable access via smart contracts — apps and AI agents can control who accesses data and trigger agentic payments autonomously

Walrus Memory leverages these properties so that agent memory is not just stored, but provably authentic and tamper-evident.

Architecture and Open-Source Lineage

The Walrus core platform is written primarily in Rust and licensed under Apache 2.0. The GitHub repository (MystenLabs/walrus) contains smart contracts for coordination and governance, Rust crates for storage nodes and clients, Docker setups, and end-to-end simulation tests. The latest release on the testnet branch was tagged testnet-v1.50.0 as of early June 2026. Walrus uses a proprietary encoding system called Red Stuff for its distributed storage, described in detail in the repository documentation.

Target Audience and Use Case

Walrus Memory is aimed at developers building AI agents that need memory persistence beyond a single session or deployment. The Walrus site frames the AI use case as: "AI is only as good as the data it's trained on. Walrus allows AI models to authenticate and verify workloads, preventing poisoned datasets or corrupted results." For agent builders specifically, Walrus Memory provides the infrastructure to store conversation history, learned context, or task state in a way that is portable across agent frameworks and verifiable by any party.

Update: testnet-v1.50.0

The Walrus GitHub repository shows active development, with the latest release tagged testnet-v1.50.0 published on June 3, 2026. The repository was last pushed to on June 5, 2026, indicating ongoing mainline development. The project has accumulated 362 stars and 118 forks on GitHub. Walrus Foundation, a not-for-profit organization, supports the ecosystem through grants and RFPs for developers building on the platform.