NemoClaw is an open-source reference stack from NVIDIA for running AI agents more safely. An AI agent is a program in which a language model plans steps and calls tools, such as a shell or a browser, to finish a task. NVIDIA describes NemoClaw as a collection of open blueprints that combine a model, an agent harness (the program that runs the model's loop of reasoning and tool calls) and a runtime into one repeatable system. The default harness is OpenClaw; Hermes and LangChain Deep Agents Code are also supported.
One shell command installs it. Each agent then runs inside an NVIDIA OpenShell sandbox, an isolated runtime that enforces which files, networks, credentials and tools the agent can reach. NemoClaw adds guided setup, managed inference that routes requests to local open models such as NVIDIA Nemotron or to cloud models, network policy, lifecycle management and a command-line tool. It can run on local NVIDIA hardware such as GeForce RTX PCs, RTX PRO workstations and DGX systems.
The code is free and open source under the Apache-2.0 license. NVIDIA calls it an alpha project.