Build an Always-On AI Assistant with OpenClaw and NemoClaw on DGX Spark

Build an always-on, fully local AI assistant on DGX Spark. This step-by-step tutorial walks through setting up a more secure, long-running AI agent using NVIDIA NemoClaw with OpenClaw for terminal interaction. It runs Nemotron 3 Super locally on DGX Spark and supports multiple interfaces and automated workflows. The result is a fully local assistant that can execute tasks, access tools, and run continuously while keeping all data and models on your own infrastructure. What you’ll build: Interfaces: Terminal, web UI, and Telegram Agent capabilities: Web research, workspace management, and task scheduling Sandboxed runtime: More secure execution with NemoClaw and NVIDIA container runtime Local model serving: Ollama-based deployment with no external dependencies 📝 Tech blog: https://developer.nvidia.com/blog/build-a-secure-always-on-local-ai-agent-with-nvidia-nemoclaw-and-openclaw/ 🛠️NemoClaw on GitHub: https://github.com/NVIDIA/NemoClaw 📑 Playbook: https://build.nvidia.com/spark/nemoclaw 🧠 Join the community on Discord: https://discord.com/channels/1019361803752456192/1482072289511211200 0:00 - Introduction to the Local AI Assistant Stack 0:21 - Interacting with the Assistant via Terminal UI 1:43 - Using the Web UI for Assistant Management 2:12 - Messaging the Agent Through Telegram 3:03 - Initial Installation and Docker Configuration 3:55 - Setting Up Ollama and Nemotron 3 Super 5:16 - NemoClaw Installation and Configuration Wizard 6:52 - Integrating Telegram, Discord, Slack 8:09 - Finalizing the Sandbox and Accessing the Dashboard 10:19 - Port Forwarding for Remote Web UI Access 12:14 - Pairing and Approving Telegram Communications