Mysten Labs has launched Walrus Memory, a new product for AI agents. The company describes it as a portable memory layer designed to allow AI agents to retain context across different applications, sessions, and model service providers, while leaving data control to the user.

Context can be preserved across multiple platforms

The company disclosed that many AI agents still rely on developers to manually assemble databases, vector storage, and runtime states, making the system prone to losing context in complex tasks. Kostas Chalkias, co-founder of Mysten Labs, stated that the main bottleneck for AI is not only computing power, but also memory capacity, which limits the agent's ability to work continuously.

Walrus Memory is designed to allow agents, applications, and workflows to share the same set of memory, rather than being tied to a single runtime environment, a single session, or a single model service provider. According to the company, this also allows multiple agents to work collaboratively in long-cycle tasks.

It has been integrated into mainstream large models.

The company states that Walrus Memory is already compatible with major AI platforms such as Claude, ChatGPT, and Gemini, aiming to reduce users' reliance on a single model provider. Developers can also integrate this memory layer into existing proxy processes via OpenClaw, NemoClaw plugins, and Python and TypeScript SDKs.

Currently, teams such as Allium, Conso Labs, Inflectiv, OpenGradient, Talus Labs, and Tatum are developing applications based on Walrus Memory, including portable agent identity systems and AI assistants that can remember customer interaction records across sessions.

Add encryption and access control

The company disclosed that the product incorporates programmable access control, allowing users to decide which models or agents can read relevant data. Mysten Labs also mentioned that the system uses cryptographic tools, including zero-knowledge proofs, for context verification and access management of encrypted memories.

Regarding performance, Chalkias stated that Walrus Memory currently focuses on optimizing the memory quality provided to large models across four key areas: storage, retrieval, sorting, and encryption. According to him, in some metrics, improved sorting, filtering, and contextual processing can lead to an improvement of approximately 60%.

Additional information:The original text bears the tag "Brought to you by Walrus," indicating it is part of a brand collaboration. The statements about performance improvements and product advantages in the text are primarily from the company.