Nvidia unveils RTX Spark software for local AI and creative tools

A shift toward local AI

Nvidia is preparing the software foundation for its upcoming RTX Spark laptops by prioritizing local AI execution. Instead of sending data to external servers, these machines are designed to run AI agents, creative applications, and coding tools directly on the hardware. This approach aims to provide a more private and responsive experience for users who want to interact with artificial intelligence without a constant internet connection.

A major part of this strategy involves making local AI agents easier for the average person to set up. When a user first turns on an RTX Spark laptop, they will see a dedicated screen to choose and install an agent app. To ensure these applications remain secure, Nvidia is incorporating OpenShell and Microsoft Execution Containers to provide sandboxing and operating system-level security.

Several popular tools are being updated to support this local-first workflow. Applications like GitHub Copilot, Perplexity, and WorkBuddy are being modified to suggest local models and settings specifically for RTX Spark systems. These updates are intended to automate the complex setup process, allowing users to get started with local AI more quickly than before.

Tools for creative professionals

Nvidia is working closely with industry leaders like Adobe to ensure that professional creative software runs efficiently on the new hardware. Adobe is currently preparing optimized versions of Photoshop, Premiere, and Substance 3D for the RTX Spark platform. These updates are expected to deliver a performance increase of as much as twofold for specific tasks involving AI, editing, color, and effects.

For video editors using Premiere, the updates focus on faster playback, rendering, and color grading. Photoshop users will see GPU-accelerated compositing, while Substance 3D Painter will receive updates to speed up texture creation and the baking of mesh maps. These improvements are designed to help creators manage large projects and complex 3D assets without the usual hardware bottlenecks.

The software ecosystem also includes specialized tools for video generation. ComfyUI is adding support for HDR and RTX Video Frame Generation, which Nvidia says will help creators handle AI-generated video within a single unified workflow. Additionally, Nvidia Broadcast 2.3 will launch with native support for the platform, allowing local AI agents to adjust settings and process files automatically.

Developer and gaming support

To support the developer community, Nvidia and Microsoft are rolling out significant updates for the Windows on Arm platform. This includes the first PyTorch-CUDA build for Windows on Arm and the first native Python build for the platform using Python 3.15. Nvidia also noted that 97% of the top 15,000 Python wheels are now ready, making it easier for developers to build and deploy software.

Gaming is another core pillar of the RTX Spark launch, with support for hundreds of titles including major multiplayer and single-player games. The platform will feature Nvidia's full graphics stack, including G-SYNC and ray tracing. Notably, DLSS 5 will arrive at launch, featuring what Nvidia calls 3D-Guided Neural Rendering to improve image quality and performance.

For the gaming community, Nvidia is also releasing specific models to enhance interactive experiences. They are launching Nemotron 3.5 Bolt E4B 4GB, a small language model with a low memory footprint designed for AI game characters and voice assistants. This allows developers to integrate smart NPCs into games without consuming the system resources needed for the main game engine.

New models and networking

Nvidia is introducing new models to power these local agents, including the Nemotron 3.5 Bolt 74B-A9B 24GB. This large language model is built specifically to run locally on RTX Spark systems and is compatible with popular tools like llama.cpp and Ollama. It is designed to handle complex tasks while remaining accessible to users who want to run high-quality AI on their own machines.

To help users manage these workloads across multiple devices, Nvidia is moving its Personal AI Router (PAIR) out of beta. This software links compatible PCs on a local network and distributes AI tasks across them. The latest version of PAIR adds support for direct model downloads from Hugging Face and custom inference settings, allowing for a more flexible way to utilize hardware resources.

The company is also introducing 'decision models' via llama.cpp to improve the speed of AI agents. These models allow agents to decide on their next action much faster than traditional systems, which is particularly useful for tasks like classification and action selection. By combining these models with the RTX Spark hardware, Nvidia aims to create a cohesive platform where gaming, creative work, and AI tasks can all happen on a single laptop.