Question 1

Can I add a GPU to a DigitalOcean Droplet for local inference?

Accepted Answer

DigitalOcean does not currently offer GPU Droplets suitable for LLM inference. Their GPU offerings are limited and not designed for sustained vLLM workloads. For cloud-based GPU inference, use CoreWeave, Lambda Labs, or RunPod, which offer H100 and A100 instances designed for AI workloads. You can run the NemoClaw sandbox on DigitalOcean and route inference to a GPU instance on a separate provider, though this adds network complexity and latency.

Question 2

Is the DigitalOcean 1-click image maintained by NVIDIA or DigitalOcean?

Accepted Answer

The 1-click image is maintained by DigitalOcean in collaboration with NVIDIA. DigitalOcean handles the image build, Ubuntu base, and marketplace listing. NVIDIA provides the NemoClaw binaries and configuration templates. Updates follow DigitalOcean's marketplace image update cycle, which typically lags 2–4 weeks behind NVIDIA's NemoClaw releases. For the latest NemoClaw version, deploy the 1-click image and then run nemoclaw update to pull the newest release.

Question 3

How does the cloud LLM API key stay secure on the Droplet?

Accepted Answer

Store the API key in /etc/nemoclaw/env with 600 permissions (owner read/write only). NemoClaw reads this file at startup and loads the key into memory. The key is never written to logs or passed through the sandbox — the privacy router makes API calls from the host process, outside the sandbox boundary. If the sandbox is compromised, the attacker cannot access the API key because it exists only in the host process's memory space, not in the sandboxed agent's environment. Never store API keys in the YAML configuration file itself.

Tag: DevOps

NemoClaw Troubleshooting: Every Known Issue and How to Fix It

Tag: DevOps

NemoClaw + DigitalOcean: 1-Click Cloud Deployment Guide

NemoClaw Troubleshooting: Every Known Issue and How to Fix It