I installed Nemoclaw due to Mr. Jensen H. security claim, but I don't know how to set it up. Please help.

[mus1u]

Hi Comunity!

Can you please give me some pointers on how to set up limitations for Nemoclaw and how to enrich it with context/skill/tools?

I have been using AI tools directly from their platforms in the past years. I did not want to download Openclaw due to security reasons.

Now that NVIDIA has put out this secure alternative, I decided to take the step and install it. Now that it is installed, I am struggling to find out how to set up appropriate limitations, add context, and experiment safely without compromising my data. I installed Qwen3 locally.

Thanks for your comments.

[nv-cyuen]

Hi mus1u,

What is your use case? There are some preset policies (i.e. Discord, Slack, Telegram) that will allow you to integrate easily as long as you have the appropriate API keys. What are you aiming to do?

[mus1u]

Hi @nv-cyuen,

Thank you for your answer. For example, I'd like to set an agent/agents to go to Chrome and access some of our proprietary software to scrap information on specific events, that otherwise i'd need to do manually, verify eligibility criteria of the events and then fill a report on a google sheets.

[nv-cyuen]

What do you mean exactly by access? Do you want the agent to go online to use the software, or view its source code like on GitHub or something?

[mus1u]

I'd like it to go online to navigate the events in the calendar of the softtware and assess if each event comply with certain criteria, then to make a report on the findings in google sheet. I hope this makes sense. Thank you for enquiring further to understand.

[nv-cyuen]

Okay got it, thanks mus1u.

So NemoClaw runs everything with a deny-by-default network access policy. So in order to connect to this proprietary software, you would need to officially approve it in the network policy (see Customize Network Policy). For your case, NemoClaw won't act like a "desktop" where it can click through your calendar like a human. If there is a way you can expose data in a structured manner, via API for example, you could then allow the API host in your policy.

Similarly with the Google Sheets aspect, you would need to allow the right Google endpoints (like sheets.googleapis.com, etc.) in your policy.

From there, you can create the appropriate scripts and skills in your sandbox to execute what you described. This is a high-level overview, and I'm happy to expand further on anything if this doesn't quite match your use-case.