The demo at CES 2026 for Nvidia's Project G-Assist is impressive in its own right. It's an AI chatbot-like interface running on the computer that can help you adjust your device's settings to fit what you're doing, with the ability to go and make those changes itself if you ask for it.

During the demo, something else about the AI tool impressed even Gerardo Delgado-Cabrera, Nvidia's director of AI PC. The demo involved asking the bot, through a microphone, to perform tasks such as adjusting the mouse sensitivity during a game. After that demonstration, Nvidia reps had to reset it to then show it to the next group. Instead of doing so manually, they simply asked the AI to do it.

Watch this: Every Announcement from the Nvidia Live CES 2026 Stream

09:02

"It's gotten to that point where we have convinced ourselves to just use it naturally," Delgado-Cabrera told me. 

Though Project G-Assist is still an experiment, you can already try it out for yourself. Nvidia has also opened it up for developers to create plugins that add functionality. It can offer recommendations while taking into account your current settings -- here's what you have, here's what could improve them. And if you ask it to make the changes, it will display the actions it's taken directly in the chat window. 

The entire process runs directly on the computer itself, rather than outsourcing the work to a data center. It will consume a significant amount of memory, so you may not want to run it constantly. 

Read more: Nvidia's Gaming Announcements at CES 2026 Are All About the Software

The same concept is behind a demo Nvidia showed of an AI chatbot assistant for the game Total War: Pharaoh. Complex strategy games like Total War can be particularly challenging for new players to grasp, given that they often come with extensive documentation and intricate mechanics that warrant their own encyclopedias. 

This AI adviser, which runs on-device rather than in the cloud, can answer the player's questions about actual in-game events using the context of all that information. I watched an Nvidia rep ask, for example, why a rebellion happened in the player's territory in the game. The adviser offered an explanation and some possible solutions.

One challenge with these kinds of AI tools is to make them helpful without taking away the fun of the game. You don't want the AI to recommend every move or spoil secrets that make a game suspenseful.

I couldn't help but imagine this interface creating a new way to interact with complicated games like Total War or Civilization. Sure, you can move the pieces around and click all the buttons like normal, but maybe you can control the game purely through voice or text chat. That is, after all, how a real-world leader does things. It wouldn't be for everyone, but it could present a totally different way to play.

Running AI models locally offers some advantages. One is privacy: Your information isn't being transferred across the internet to the cloud, over connections that may or may not be as secure as you'd like. Another is speed, in that these models take a lot less time to work because they're running right there on local chips. 

But for power users like designers and creators, all the different iterations of AI generation required to tweak something and make it perfect can add up to a lot of money from a cloud service. Run it on your own device, and you're saving on potentially huge subscription and usage costs.