NVIDIA at Gamescom 2025 announced a major update to Project G-Assist, its locally accelerated AI chatbot and utility. The company developed a new lighter AI model for G-Assist, which can make do with as little as 6 GB of video memory. It earlier needed 12 GB or higher. The move to lower video memory requirement opens the doors to 11 GB cards such as the RTX 2080 Ti, 10 GB cards such as the RTX 3080, a plethora of cards with 8 GB, and also 6 GB ones such as the RTX 2060.

This is achieved from a combination of the AI model having 40% lower memory footprint, and improved tool-calling intelligence. This should prove particularly useful for certain laptop GPU models that come with lower video memory sizes than their desktop counterparts. G-Assist can be invoked by gamers in game to run diagnostics or optimize game performance, provide graphics sub-system information and monitoring, and adjust GPU or gaming peripheral settings, using simple natural language prompts. NVIDIA also announced the G-Assist plug-in hub in collaboration with Mod.io, which lets users add plug-ins for things such as game peripherals. The company recently ran the G-Assist plug-in hackathon, and put out some winning project examples, such as Omniplay, Launchpad, and Flux NIM microservice for G-Assist.