What it does
NemoClaw is an open source reference stack from NVIDIA for running supported AI agents inside OpenShell sandboxes. The agents it supports are OpenClaw (the default), Hermes and LangChain Deep Agents Code. It provides guided onboarding, managed inference, network policy, managed integrations and state that persists in OpenShell. Day-to-day work happens through the NemoClaw command-line tool and agent-specific aliases of it. The project describes itself as alpha. Maintainers review issues and pull requests on a best-effort basis with no guaranteed response times, and its stated priorities are install reliability, sandbox hardening, credential handling, network-policy defaults and checking local and routed inference.
Capabilities
- Runs OpenClaw, Hermes or LangChain Deep Agents Code inside an OpenShell sandbox
- Guided onboarding, with an express install for supported DGX or WSL hosts
- Managed inference with a choice of inference provider
- Network policy with baseline rules, operator approval and egress control
- Sandbox hardening such as capability drops and process limits
- Lifecycle operations through the NemoClaw CLI
When to use it
- You want to run an agent like OpenClaw with network and credential controls around it
- You need a reference setup for sandboxed agents on NVIDIA OpenShell
- You are comfortable with an alpha project and best-effort support
