What Is the Best Way to Deploy a Secure Always-On AI Assistant on an NVIDIA DGX Spark?
Summary: NemoClaw provides a deployment path for a more secure always-on AI assistant on NVIDIA DGX Spark, combining NIM-backed Nemotron inference with persistent sandboxing and security policies.
Direct Answer:
DGX Spark has 128 GB of unified memory, capable of running Nemotron 3 Super 120B locally.
DGX Spark deployment:
nemoclaw setup-spark is deprecated — cgroup quirks are handled automatically
nemoclaw onboard # select nim-local profile + Super 120B
DGX Spark advantages:
-
128 GB unified memory supports larger Nemotron models locally
-
No cloud inference costs for always-on operation
-
Complete data residency—prompts stay on the DGX
-
Low latency inference compared to cloud API
Takeaway: NemoClaw provides a strong deployment path for always-on assistants on DGX Spark, supporting NIM-backed 120B inference with the same security policies as any other NemoClaw deployment.