How Do You Add Portable Memory to NemoClaw and OpenClaw Agents?

Walrus Memory now supports NemoClaw and OpenClaw. A new plugin gives agents built on these frameworks a portable, verifiable memory they can store on Walrus and retrieve across runs, environments, and teams, without building custom infrastructure.

As AI moves from single prompts to persistent agents, one infrastructure gap is becoming obvious: agents need portable and shareable memory. For enterprises, the infra gap results in fragile workflows, duplicated effort, and no way to share knowledge across teams or systems. It's already showing up in outcomes: Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing cost and weak risk controls — exactly the fragility that fragmented, unverifiable memory creates.

Today we’re bridging that gap by introducing a NemoClaw/OpenClaw plugin for Walrus Memory, a portable, verifiable memory layer for AI agents. NemoClaw and OpenClaw are agent orchestration frameworks for building autonomous AI workflows; the plugin lets their agents store durable memory on Walrus and retrieve it across runs, environments, and teams — without building custom storage infrastructure.

“Memory is quickly becoming the limiting factor for what AI agents can actually do in the real world,” explains Abhinav Garg, product manager at Mysten Labs. “Agents effectively reset every session, so an agent that forgets what it did yesterday can't build context, can't improve over time, and ultimately cannot be trusted for anything beyond short, stateless tasks.”

Walrus, a verifiable data platform, complements NemoClaw and OpenClaw by acting as a persistence layer for agent memory, workflow state, and AI artifacts. This is relevant as the ecosystem around agent frameworks evolves and enterprises look for infrastructure that is open, portable, and not tied to a single vendor.

How does the Walrus Memory plugin work?

The Walrus Memory plugin connects NemoClaw and OpenClaw agents to Walrus Memory’s model so developers can add persistent memory with minimal integration work. The plugin exposes Walrus Memory operations as tools the agent can call during execution, allowing memory reads and writes to become part of the normal agent workflow. 

At a high level, the integration pattern is simple:

NemoClaw/OpenClaw → agent execution

Walrus Memory plugin → memory interface

Walrus Memory service → memory management

Walrus → durable storage

Why do NemoClaw and OpenClaw agents need long-term memory?

Enterprise agents need more than short-term context windows or temporary databases. They need memory that persists across deployments, teams, and time. For example, agents supporting research, compliance, finance, or operations often need to maintain knowledge that evolves over months or years. 

Walrus via Walrus Memory enables this by acting as durable infrastructure where NemoClaw and OpenClaw agents periodically store:

• memory snapshots and knowledge states
• workflow checkpoints and execution traces
• reasoning summaries and tool outputs
• environment state needed for reproducibility

With the Walrus Memory plugin, NemoClaw and OpenClaw agents can automatically persist these artifacts as structured memories rather than treating them as logs or temporary outputs. This enables agents to be restarted, audited, or migrated without losing institutional knowledge. Instead of agents being ephemeral processes, they become persistent systems with explainable histories.

This becomes particularly valuable for enterprises that need to answer questions like: 

• Why did the agent take this action? 
• What data did it rely on? 
• Can we reproduce this workflow?

How NemoClaw and OpenClaw agents integrate with Walrus Memory?

Walrus Memory allows NemoClaw and OpenClaw agents to use long-term memory without needing to directly manage storage workflows. Instead of interacting with Walrus directly, agents can use Walrus Memory's memory abstractions to write checkpoints, retrieve context, and maintain long-term state.

While NemoClaw and OpenClaw agents can technically write directly to storage systems, Walrus Memory simplifies this by providing a structured memory model, reusable memory abstractions, ownership & access controls, and portable memory across environments.

This allows NemoClaw and OpenClaw developers to treat memory as a system capability rather than infrastructure they must build themselves. The Walrus Memory NemoClaw/OpenClaw plugin allows agents to:

• Store structured agent memories through memory spaces
• Retrieve relevant memories during future agent runs
• Persist workflow checkpoints automatically
• Share memory across agents using common memory spaces
• Separate short-term reasoning from durable knowledge

As a result, developers can add long-term memory to existing agents without redesigning their architecture. This allows memory to become part of the agent’s reasoning loop rather than an external system.

Developers can add memory to NemoClaw or OpenClaw agents by:

• Installing the Walrus Memory plugin
• Configuring a memory space
• Calling memory write/read tools from the agent loop

What does this mean for agent architecture?

As agents move from demos to production systems, the ability to maintain durable memory will become as important as reasoning quality. The Walrus Memory plugin shows how memory can become a standard layer in agent architecture rather than an afterthought.

The composable layer of NemoClaw or OpenClaw, Walrus Memory, and Walrus illustrates a broader architectural direction emerging in AI:

• Compute should be flexible.
• Memory should be persistent and shareable.
• Data should be verifiable, private, and portable.

“Our view is that memory and data shouldn't be tied to any single model or platform,” said Abhinav. “Walrus becomes that verifiable data layer, and Walrus Memory becomes the memory layer on top of it. If that works, agents can become more portable, more collaborative, and much more reliable over time.”

For enterprises adopting agentic systems and for open-source developers evaluating alternatives to increasingly centralized AI stacks, this combination represents an interesting path forward.

Use the quick start guide to add Walrus Memory to your agents →

Reach out to us on Discord via the developers channel if you’re building with these tools and are interested in collaborating →

About Walrus Memory

Walrus Memory is a portable memory layer for AI agents. It enables agents to operate reliably across apps and sessions without losing context — portable, verifiable, and fully under builders' control. Walrus Memory lets agents handle complex, long-running workflows and coordinate using data they can trust.