Nvidia RTX Spark Laptops Launch to Run Local AI Agents on Windows
New Nvidia-powered PCs feature Blackwell chips and Grace processors, enabling local execution of AI agents like GitHub Copilot and Hermes.
Key points
- RTX Spark laptops combine Blackwell GPU and Grace CPU in a single superchip with up to 128GB shared memory.
- Users can install AI agents such as OpenClaw and Perplexity directly from the first startup without using a command line.
- Nvidia released Nemotron 3.5 Bolt, a 74-billion-parameter model optimized for these local configurations.
Nvidia launched its RTX Spark laptops on Friday, introducing a new class of Windows devices designed to run artificial intelligence agents locally. According to The Next Web, the systems allow users to install tools like GitHub Copilot, Hermes, OpenClaw, and Perplexity immediately upon first startup, eliminating the need for command-line configuration.
Each device operates on a single superchip that pairs a Blackwell RTX graphics chip with a Grace processor. This architecture provides up to 128GB of shared memory and delivers up to one petaflop of AI performance. MediaTek collaborated with Nvidia on the processor design, enabling the hardware to handle large language models and autonomous agents entirely on the device.
Hardware and Software Integration
Jensen Huang, Nvidia’s founder and chief executive, stated that RTX Spark combines NVIDIA AI, RTX graphics, and CUDA with Microsoft Windows for creating, gaming, and running personal AI agents. Yusuf Mehdi, executive vice president at Microsoft, said the company is building Windows as a platform for hybrid intelligence, allowing AI to run locally when it makes sense and reach the cloud when needed.
Nvidia released Nemotron 3.5 Bolt on Thursday to support these devices. The model contains 74 billion parameters, with nine billion active at any given time, compressed into 24GB of memory. Nvidia confirmed that the model’s weights, training data, and recipes are open source and compatible with llama.cpp, Ollama, LM Studio, Unsloth, and vLLM.
Security and Ecosystem Support
To manage agent permissions, Nvidia introduced the OpenShell runtime alongside Microsoft Execution Containers. These tools let users control what actions agents are permitted to perform. OpenClaw and GitHub Copilot already support Microsoft’s containers, while Hermes will add OpenShell controls by the end of October, according to a blog post from Nvidia.
Dillon Rolnick, chief executive of Nous Research, noted that bringing Hermes to RTX Spark allows users to leverage open AI models locally on Windows laptops. This approach offers greater control over data and the freedom to customize AI workflows. Additionally, Adobe announced that Photoshop, Premiere, and Substance 3D are optimized for the chip, with AI processing and editing speeds up to twice as fast.
What should you do?
Professionals evaluating these systems should verify that their organization’s security policies permit local AI agent execution before deployment. Users should review the permission scopes granted to agents like Copilot or Hermes via the Microsoft Execution Containers. Organizations should monitor network traffic for unexpected cloud fallbacks when using the Switchyard model router. IT teams should ensure that endpoints running OpenShell are kept up to date with the latest runtime patches.
Background: Large language models
Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.
What to do and how to stay safe: Nvidia
- Review local AI agent permissions using Microsoft Execution Containers to restrict unauthorized actions.
- Monitor system resources to ensure the 128GB shared memory is not exhausted by large local models.
- Verify that third-party agents like Hermes support the latest OpenShell security controls before installation.
- Assess whether local processing meets data privacy requirements compared to cloud-based alternatives.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Which AI agents are pre-installed or available for RTX Spark laptops?
Users can install GitHub Copilot, Hermes, OpenClaw, and Perplexity from the first startup without using a command line.
What is the memory configuration of the RTX Spark superchip?
The superchip pairs a Blackwell RTX graphics chip with a Grace processor and offers up to 128GB of shared memory.
How does Nvidia manage security for local AI agents?
Nvidia uses the OpenShell runtime and Microsoft Execution Containers to control what actions agents are allowed to perform.
