The tutorial is based on the open-source NemoClaw stack, integrating OpenShell and OpenClaw, providing a solution for long-term, secure deployment of autonomous AI assistants locally. Using the NVIDIA DGX Spark system as an example, it details the environment requirements (Ubuntu 24.04, Docker 28.x, Ollama, Telegram bot token, etc.) and deployment steps, which are expected to take 20–30 minutes, with an additional 15–30 minutes to download approximately 87GB of models. The core components include NemoClaw, OpenShell, OpenClaw, Nemotron 3 Super 120B, and NIM/Ollama inference deployment. The article notes that although there is strong isolation, no sandbox can fully prevent advanced prompt injection; testing should be conducted on isolated systems.

MeNews

2026-05-25 09:36:18

Abstract generation in progress

ME News Report, April 18 (UTC+8), NVIDIA recently released a technical tutorial guiding developers on how to build a secure, long-running, fully local autonomous AI assistant. The tutorial is based on NVIDIA's open-source reference stack NemoClaw, which integrates OpenShell secure runtime and OpenClaw self-hosted gateway, aiming to address data privacy and control risks when deploying AI agents on third-party clouds. The tutorial provides detailed deployment steps on NVIDIA DGX Spark (GB10) systems, including environment setup, local service models, installation stack, and connecting to Telegram. Deployment requires meeting specific hardware (DGX Spark running Ubuntu 24.04 LTS), software (Docker 28.x+, Ollama), and prerequisites such as creating a Telegram bot token. Estimated active operation time is 20-30 minutes, plus an initial model download of about 87 GB taking 15-30 minutes. Core components include NemoClaw, OpenShell, OpenClaw, Nemotron 3 Super 120B LLM, and inference deployment with NIM or Ollama. The article also notes that while OpenShell offers strong isolation features, no sandbox can provide complete protection against advanced prompt injections, and it is recommended to deploy new tools on isolated systems during testing. (Source: InFoQ)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

10 Likes

Reward
10
7
2
Share

Comment

Add a comment

Half-SectionedSucculent

· 6h ago

Isolation system testing is a painful lesson; previously experimenting with LLM almost caused the main environment to collapse.

View OriginalReply0

FomoLibrarian

· 6h ago

Can it run in just 20 minutes? Faster than I can set up a node.

View OriginalReply0

ReefUnderTheMoonlight

· 6h ago

NIM/Ollama dual inference options, maximum flexibility

View OriginalReply0

DaoPeripheralWorker

· 6h ago

The biggest advantage of an open-source stack is that it can be customized, waiting for a community to fork.

View OriginalReply0

FloatingMirrorSphere

· 6h ago

Ubuntu 24.04+LTS，版本控狂喜

Reply0

CapitalFlowInATeacup

· 6h ago

Running 120B parameters locally, if quantized, can it still perform well?

View OriginalReply0

ViewingNarrativesFromAHotAir

· 6h ago

With the 87GB model, my hands are sore, and the hard drive is burning.

View OriginalReply0

Trending Topics
View More
#
StockTradingChallengeUpTo17000U
15.93M Popularity
#
USIranDraftDeal
288.55K Popularity
#
TradeCFDWinGold
3.03M Popularity
#
HYPEMarketCapSurpassesDOGE
12.64M Popularity
#
PlatinumCardCreatorExclusive
155.83K Popularity

Pinned

Sitemap

NVIDIA releases tutorial on building a local sandboxed AI assistant based on NemoClaw

Trending Topics

StockTradingChallengeUpTo17000U

USIranDraftDeal

TradeCFDWinGold

HYPEMarketCapSurpassesDOGE

PlatinumCardCreatorExclusive

Pinned