Walrus Memory Wants Agents to Carry Their Memory Between Apps

Walrus launched Walrus Memory on June 3, pitching it as a portable, verifiable memory layer built specifically for agents. The promise is that an agent's context survives across apps and across sessions, and stays under the builder's control rather than getting trapped inside whatever product it happened to run in.

It ships with day-one support for Claude, ChatGPT, and Gemini, plus the OpenClaw and NemoClaw agentic frameworks. The angle that separates it from the pack is portability and verifiability, with roots in decentralized storage, so the memory is not just a vendor's database you are renting but something you can move and audit.

What makes this worth flagging is the pattern around it. Walrus Memory is the third agent-memory product to land in roughly a week, after Supermemory on June 1 and Hyper on June 4. When three teams ship the same category in seven days, that is the market telling you something. Agent memory is graduating from a thing every team rebuilds by hand into a layer you buy off the shelf.

The open question is whether portability is the wedge that wins, or whether the model providers just absorb memory into their own platforms and the standalone layer gets squeezed. Either way, the buy-versus-build line for agent memory just moved. https://walrus.xyz/products/walrus-memory/