What problem does it solve?
Running NemoClaw on remote GPU instances requires coordinating provisioning, installer versus legacy deploy flows, onboarding for OpenShell-managed messaging channels, and a set of image- and runtime-level hardening controls so agents remain sandboxed and credentials are handled safely. This Skill documents the end-to-end steps, warnings, and configuration knobs operators need to deploy a sandboxed OpenClaw agent remotely while preserving least-privilege and reproducible credential rotation.
Core Features & Use Cases
- Provisioning guidance: Deprecated Brev compatibility path and the recommended pattern of provisioning a host, running the NemoClaw installer there, and completing the onboard flow.
- Security hardening: Image and runtime controls including process limits, capability dropping, Landlock/read-only home policies, and notes about kernel requirements and fallback behavior.
- Messaging and onboarding: How Telegram, Discord, and Slack tokens are registered with OpenShell providers, how channel config is baked at image-build time, and how to rotate credentials safely.
- Use case: Deploy a GPU VM for a research assistant, configure CHAT_UI_URL and proxy settings for remote dashboard access, onboard the sandbox with a Telegram bot token, and verify inference and network policy from the host.
Quick Start
Provision a remote GPU host, install NemoClaw on that host, and run the onboarding wizard on the host to build and launch a hardened sandbox with your messaging channels configured.