NVIDIA NemoClaw
A phone line for your NemoClaw agent.
NemoClaw runs an OpenClaw agent inside NVIDIA's OpenShell sandbox. Our voice gateway connects to OpenClaw agents through their Responses endpoint, so that is the way in. There is no packaged NemoClaw install yet: we set it up with you.
How the pieces fit. And what your policy allows.
NemoClaw manages the sandbox, its isolation and its network policy. Inside it is an OpenClaw agent. The gateway runs next to it and makes the same connection it makes to any OpenClaw agent.
OpenShell blocks outside access until it is approved, so the policy has to allow the gateway's connection to the agent and the network traffic a phone call needs.
- The agentThe OpenClaw agent inside your NemoClaw sandbox
- The connectionThe gateway calls OpenClaw's /v1/responses over loopback
- Your policyApproves that connection and the call's outbound traffic
- NetworkThe gateway opens no inbound ports
- StatusNo NemoClaw installer profile or policy templates yet. Set up together
What leaves your environment. Stated plainly.
Teams choose NemoClaw to control where data goes. A call through us cannot stay entirely on the device, so here is exactly what moves.
Stays with your agent
Its tools, memory and data. It keeps using whichever model you configured, local or not.
Leaves for the call
Our carrier connects the call and an encrypted tunnel carries it to the gateway. Google's realtime voice model does the listening and speaking, and receives the call audio plus the context it needs to answer: the agent's instructions and what its tools return.
What we keep
Recordings and transcripts are separate switches, both on by default. Call details such as numbers, duration and timing are always kept for your dashboard.
If audio from a call must never leave your network, this design is not a fit today. We'd rather say so up front.
On a call. The OpenClaw agent you built.
Your skills and tools
Mid-call, the gateway asks the sandboxed agent to look something up or take an action, and speaks the answer.
Knowledge base search
Answers from the documents and data your agent already has.
Inbound and outbound
Answers your number, and places calls through the Voice API.
Questions. Straight answers.
Is NemoClaw supported today?
Not as a packaged install. NemoClaw runs OpenClaw, and the gateway connects to OpenClaw through a documented endpoint, so the path is clear. But there is no NemoClaw installer profile and no ready-made OpenShell policy yet. We work through the setup with you, for your environment.
How is NemoClaw different from OpenClaw?
Same agent, different containment. OpenClaw is the open-source agent framework. NemoClaw is NVIDIA's stack that runs OpenClaw inside an OpenShell sandbox, with isolation and policy approval for outside access.
Does the call stay on the device?
No. The agent's tools and data do. The call audio travels through our carrier and to Google's realtime voice model, which does the listening and speaking. If that does not meet your requirements, a phone line from us is not the right fit yet.
What should I prepare?
A running NemoClaw stack with its OpenClaw agent, access to the host, and someone who can approve changes to the sandbox policy. Then tell us about your setup.
Is there a simpler way to try it?
Yes. Connect a phone line to a plain OpenClaw agent first. It is a single install command, and the experience on the call is the same.
Running NemoClaw? Let's talk it through.
Tell us about your environment and your sandbox policy. We'll tell you plainly what a phone line for your agent involves.