Top AI Product

AI agents forget. Close the session and everything the agent learned about you resets to zero. Walrus Memory, launched June 3, is a memory layer built to fix that — agents keep context, share knowledge, and build on past work across sessions, apps, and runtimes.

## How it works

Developers wire in two calls: remember and recall. That’s the whole API. It ships with direct support for Claude, ChatGPT, and Gemini, plus the OpenClaw and NemoClaw agent frameworks, so an agent can pull what it learned in one app into another instead of starting cold every time.

## The verifiable angle

The twist is control. Walrus Memory runs on Walrus, a verifiable data platform, and every memory is tied to a signing key you hold — only signed requests can read or write it, with programmable permissions deciding which agents touch which memories. That’s a different bet from the usual vector-database memory: portable and user-owned, not locked inside one app’s backend.