Take your agent’s memory anywhere

Walrus Memory is a portable memory layer that makes AI agents reliable across apps and sessions. Persistent, verifiable, and fully under your control.

Start building
// npm install @mysten-incubation/memwal
// pnpm add @mysten-incubation/memwal
// yarn add @mysten-incubation/memwal
import { MemWal } from "@mysten-incubation/memwal";
const walrusMemory = MemWal.create({
key: "<your-ed25519-private-key>",
accountId: "<your-memwal-account-id>",
serverUrl: "https://your-relayer-url.com",
namespace: "demo",
});
await walrusMemory.health();
const job = await walrusMemory.remember("I live in Hanoi and prefer dark mode.");
await walrusMemory.waitForRememberJob(job.job_id);
const result = await walrusMemory.recall({ query: "What do we know about this user?" });
console.log(result.results);

Builders shipping
with Walrus Memory

Conso

Without memory,
agents hit reset

Walrus lying down

State isn't shared. Agents make conflicting decisions.

Memory is stale or unverifiable. Output isn't reliable in production.

Something breaks. No audit trail, no way to trace what your agent acted on.

A portable memory layer for AI agents

Portable by design

Context doesn't die when the session ends. Your agent picks up where it left off — in a different app, in a different runtime, weeks later.

Your agents keep context wherever they run

Yours to control

You decide how every memory is shared, accessed, and updated. Programmable permissions keep privacy on your terms.

Explicit ownership and access control

Built for coordination

In multi-step workflows, your agents share memory and stay coordinated on the same state. Memory integrity is independently verifiable.

Multiple agents, one source of truth

Plugs into the stack you already have

Walrus Memory ships with native support for the platforms and protocols teams are already building on.

  • Works with every major LLM out of the box

  • First-party plugins for OpenClaw and NemoClaw

  • Native MCP support. No adapters needed

  • Python, TypeScript, and JavaScript. Pick your language

What builders are creating with Walrus Memory

A portable agent memory layer doesn’t just improve your stack. It changes how your product behaves — and what users come to rely on.

Multi-agent workflows

Agents share context across tasks, tools, and time. What one agent learns can become available to the entire workflow.

Customer support agents

Pick up where the last conversation ended. Know what was tried, what failed, and what the customer actually needs.

Notes and research apps

A second brain that compounds over time. It captures ideas continuously and retrieves them naturally weeks later.

Personal assistants

Remember tone, preferences, routines, and relationships. Feel like assistants that know you, not new hires every session.

Memory that holds up in production

Context travels with the workflow: programmable permissions, verifiable state, no lock-in. Built on Walrus, the Verifiable Data Platform.

Remember

Authenticate

Your SDK sends a signed request so the relayer can verify ownership and access.

Process

The relayer analyzes and encodes your memory using embedding models optimized for retrieval.

Encrypt & store

Your memory is encrypted, stored on Walrus, and indexed for fast semantic search.

Walrus Memory mascot

Recall

Authenticate

Your SDK sends a signed query request so the relayer can verify access permissions.

Search

The relayer interprets the query, retrieves the most relevant memories, and fetches them from Walrus.

Decrypt & inject

The memory is decrypted and injected into the model context for downstream reasoning.

Build agents
that remember

Wire memory into your agent.
Ship without rebuilding context every session.

FAQs