Give NVIDIA NemoClaw a voice and a phone number
NemoClaw runs OpenClaw inside NVIDIA's secure, on-device sandbox. Talk To My Agent connects to that same OpenClaw core to add a phone number and a natural voice - so even regulated teams can put their agent on the phone without data leaving the box.
The short answer
Yes - you can give a NemoClaw agent a phone number and a natural voice with Talk To My Agent (TTMA). NemoClaw is NVIDIA's enterprise stack that runs an OpenClaw agent inside an OpenShell sandbox with network isolation and local inference. Because NemoClaw runs OpenClaw, TTMA connects to that same local OpenClaw gateway and adds telephony and voice - on the same device, with the call's external (carrier) leg governed by NemoClaw's policy approval. The result is a compliant phone line for your agent where no agent data leaves your environment.
NemoClaw + Talk To My Agent - at a glance
Enterprise · via OpenClaw core- Agent stack
- NVIDIA NemoClaw (OpenShell sandbox running OpenClaw)
- How it connects
- TTMA voice gateway → the sandboxed OpenClaw
/v1/responses, on-device - Data residency
- On-device - agent reasoning and data stay in the sandbox
- External access
- Telephony egress governed by OpenShell policy approval
- Voice
- Natural, human-sounding · pick your voice · 20+ languages
- Setup
- Enterprise - connect via the OpenClaw core; talk to us
What is NVIDIA NemoClaw?
NemoClaw is NVIDIA's open-source reference stack for running AI agents securely. In NVIDIA's words, it “orchestrates NVIDIA OpenShell to run OpenClaw” - the self-hosted agent gateway - inside a hardened sandbox. OpenClaw lives in the sandbox; NemoClaw manages its lifecycle, isolation, and policy.
The point is data governance. NemoClaw adds network and filesystem isolation, real-time policy approval for any external access, and full local inference, so “no data leaves the device during agent operation.” That makes it a fit for healthcare, finance, government, insurance, and legal - regulated places a public-cloud agent can't go.
Why give NemoClaw a phone number?
Regulated teams want an agent on the phone too - but only if it can't leak data. Because the voice gateway runs on-device next to the sandboxed agent, NemoClaw is exactly where a compliant phone line makes sense.
- Compliant by design. The agent's reasoning and data stay inside the sandbox; only the carrier leg leaves, under policy approval.
- On-device, local inference. Fits NemoClaw's “no data leaves the device” model - the voice gateway is local too.
- A phone line for regulated work. Patient scheduling, member services, case intake - answered in a human voice, 24/7.
- Same OpenClaw agent. The skills, memory, and knowledge base you've built run on the call - now governed by NemoClaw.
- You control egress. OpenShell's real-time policy approval decides what the call can reach.
How to give NemoClaw a phone number
NemoClaw runs OpenClaw, so the phone line connects through that OpenClaw core - inside the sandbox, with egress under your policy.
- 1
Run your NemoClaw stack
Stand up NemoClaw so it orchestrates OpenShell to run your OpenClaw agent in its sandbox (see NVIDIA's guide). The OpenClaw gateway is reachable on the sandbox's loopback.
- 2
Install the TTMA voice gateway on the same host
Bring up TTMA's gateway alongside the stack so it talks to the sandboxed OpenClaw
/v1/responsesover loopback - the same connection TTMA uses for any OpenClaw agent, so the agent's data never leaves the device.curl -sSL https://api.talktomyagent.io/install.sh | bash -s -- --token <YOUR_TOKEN> --agent-profile openclaw --accept-license - 3
Allow the connection in OpenShell policy
OpenShell blocks external access until approved. Allow the voice gateway's local connection to the OpenClaw gateway, and approve the telephony egress your calls require - so you decide exactly what the phone line can reach.
- 4
Pick a voice, number, and publish
Choose a voice and a local number, then publish it. Your NemoClaw-governed agent now answers and places calls in a natural voice.
Enterprise note: NemoClaw runs OpenClaw, so TTMA connects through the OpenClaw core today. A dedicated NemoClaw installer profile and OpenShell policy templates are on our roadmap - contact us and we'll help you wire it up for your environment.
Talk to us about enterprise / NemoClaw setupWhat your NemoClaw agent can do on a call
The full OpenClaw agent on the phone - inside NemoClaw's guardrails.
Keeps data on-device
Agent reasoning and data stay in the sandbox; only the carrier leg leaves, under policy.
Uses your skills & tools
The sandboxed OpenClaw agent runs its skills and tools live on the call.
Searches your knowledge base
Answers from the internal knowledge base your agent already has.
Egress under policy
OpenShell approves or denies what the call can reach, in real time.
Speaks 20+ languages
Auto-matches the caller's language on the fly.
Crystal-clear audio
Built-in noise cleanup keeps the line clear on both ends.
Frequently asked questions
What is NVIDIA NemoClaw?+
NemoClaw is NVIDIA's open-source reference stack for running AI agents securely. It orchestrates NVIDIA OpenShell to run an OpenClaw agent inside a hardened sandbox with network and filesystem isolation, real-time policy approval, and local inference - so no data leaves the device. It's aimed at regulated industries like healthcare, finance, government, and legal.
Can a NemoClaw agent make and receive phone calls?+
Yes - with Talk To My Agent. Because NemoClaw runs OpenClaw, TTMA's voice gateway connects to that same OpenClaw core on-device and adds a phone number and a natural voice, so the agent can answer and place calls.
Does giving NemoClaw a phone keep data on-device?+
Yes. The voice gateway runs locally and talks to the sandboxed OpenClaw agent over loopback, so the agent's reasoning and data stay inside your environment. The only thing that leaves the box is the carrier (telephony) leg of the call, which OpenShell's policy approval governs.
How is NemoClaw different from OpenClaw?+
Same agent, different containment. OpenClaw is the open-source agent framework; NemoClaw is NVIDIA's enterprise stack that runs OpenClaw inside an OpenShell sandbox and adds isolation, policy approval, and local inference for regulated environments. TTMA gives either one a voice - for NemoClaw, through its OpenClaw core.
Is NemoClaw voice generally available?+
TTMA connects to NemoClaw today through its OpenClaw core - the same path TTMA uses for any OpenClaw agent. A dedicated NemoClaw installer profile and OpenShell policy templates are on our roadmap. For a guided enterprise setup, contact us.
Which industries is this for?+
Anywhere regulated data flows through AI - healthcare, financial services, government, insurance, and legal - that wants an agent on the phone without sending data to a public cloud.
Give any agent a phone number
A compliant phone line for your NemoClaw agent
Give your NVIDIA NemoClaw agent a number and a natural voice - on-device, with egress under your policy. Let's set it up for your environment.