NeuralStackNVIDIA RTX Spark laptops become available today, Oct. 16, as NVIDIA and its software partners position the platform as a new type of Windows PC. The system combines gaming, creative production and local AI in one device, with up to 128GB of unified memory and 1 petaflop of AI compute. NVIDIA says thousands of apps, games and models are ready for launch.
The launch centers on AI agents that can run locally while still connecting to cloud models when needed. Updates to GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy are intended to simplify setup on RTX Spark, improve security and support hybrid local-and-cloud routing. NVIDIA Switchyard, an open-source model router library, is designed to help agents choose between local and cloud language models based on the task.
Security is another major part of the rollout. NVIDIA OpenShell and Microsoft Execution Containers, or MXC, provide sandboxing and operating-system-level protections for agents, while OpenShell adds credential protection and security monitoring. OpenClaw and GitHub Copilot support MXC, and Hermes Agent is expected to add more granular OpenShell controls by the end of October.
NVIDIA is also introducing the Nemotron 3.5 Bolt 74B-A9B model for local agents on RTX Spark. The mixture-of-experts model contains 74 billion parameters, activates 9 billion parameters and is packaged into 24GB using 2-bit quantization. NVIDIA says it is designed to balance speed and intelligence for agentic workloads running locally.
Creative software is receiving its own round of RTX Spark optimizations. NVIDIA says performance improvements of up to 2x are coming soon to Adobe’s top applications, following work to rearchitect Photoshop compositing and improve Premiere playback, rendering and effects. NVIDIA Broadcast 2.3 adds Video Super Resolution, Video Frame Generation, offline file processing and controls for compatible agents.
Developers can use PyTorch, Python, CUDA and CUDA libraries on the platform, while Llama.cpp adds decision-model support and a Windows app. NVIDIA PAIR is leaving beta with Llama.cpp support, and NVIDIA is releasing application-specific Model Context Protocol tools along with AI playbooks for tasks such as building a chief of staff, configuring a development system or creating a design assistant.
Gaming remains part of the launch strategy, with hundreds of supported games and anti-cheat systems. A faster DLSS 5 AI model arrives today, while a Performance Mode for DLSS Frame Generation is available through the NVIDIA App beta; Nemotron 3.5 Bolt E4B for NVIDIA ACE game characters and an RTX Remix update with DLSS 5 are scheduled for Oct. 23.
Why this matters
RTX Spark reflects a wider effort to make local AI agents practical for everyday PC users rather than limiting them to cloud services or specialist development machines. Easier installation, sandboxing and automatic model selection address some of the friction that has slowed adoption, while hybrid routing lets users preserve local workflows without giving up access to more capable cloud systems. If the software ecosystem delivers on NVIDIA’s launch promises, the PC could become a more active platform for personal AI, creative work and development.
NVIDIA’s next challenge will be sustaining that momentum beyond launch. The company is relying on continued support from application developers, model builders, game publishers and security providers, while upcoming software releases will determine how quickly RTX Spark’s local-agent vision becomes a routine part of Windows computing.