NVIDIA readies local AI apps for RTX Spark laptops arriving October 16

Running an AI assistant on a laptop is getting a more guided setup with NVIDIA’s RTX Spark rollout. In an October 13 update, NVIDIA detailed agent installers, local-model support and developer tools ahead of the laptops’ scheduled October 16, 2026 availability. The software spans ready-to-use tools, beta features and integrations still due later this month.

From first boot to a working agent

NVIDIA says RTX Spark’s first-start screen will let users choose and install an agent app. GitHub Copilot, Hermes Agent, OpenClaw, Perplexity, Pi and WorkBuddy will recommend an optimized local model and settings, then handle setup in a few clicks. That puts model selection and configuration inside the app rather than leaving users to assemble the pieces themselves.

Security support has its own timetable. Microsoft Execution Containers, or MXC, supplies operating-system-level boundaries for agents. NVIDIA’s OpenShell helps agents use those containers and adds credential protection and security monitoring. OpenClaw and GitHub Copilot already support MXC; Hermes Agent is scheduled to integrate MXC and more detailed OpenShell controls by the end of October.

Local execution is not the only route. NVIDIA’s open-source Switchyard library lets agents route requests between local and cloud models. The company presents it as a way to reduce cloud use while retaining access to stronger remote models. Hermes is due to add Switchyard later in October; Perplexity Computer and GitHub Copilot have their own routing systems.

A local model packed into 24GB

The accompanying Nemotron 3.5 Bolt 74B-A9B model is designed for local agents on RTX Spark. NVIDIA describes a mixture-of-experts design with 74 billion parameters, of which 9 billion are active. It uses 2-bit quantization—a lower-precision representation of the model’s numbers—to fit into 24GB. NVIDIA says its training approach is designed to preserve quality despite that compression.

Users will be able to run it through llama.cpp, vLLM, Ollama, Unsloth and LM Studio, with direct access through agents including Hermes and OpenClaw. NVIDIA also says it will release training data, model weights and recipes, giving developers material to customize their agents rather than only a finished model to run.

Development tools arrive; creative updates remain phased

For developers, the rollout includes native Windows on Arm tools, not just agent apps. NVIDIA identifies three additions arriving alongside RTX Spark:

• A PyTorch-CUDA build for Windows on Arm.
• Python 3.15, with a build native by default on Windows on Arm.
• A native Conda installer, with more than 2,000 packages.

NVIDIA says 97% of the top 15,000 Python wheels—installable software packages—are ready, with work continuing on the remainder. Developers can use Windows directly or Windows Subsystem for Linux to access packages used for cloud deployment.

Creative-app readiness is less uniform. RTX Spark-optimized AI features in Photoshop and Premiere are available in beta, while Adobe Substance 3D updates are still coming. NVIDIA claims up to twice the performance across Adobe creative workflows, but says the updates will roll out in phases.

The rollout also extends beyond one laptop. PAIR, NVIDIA’s free, open-source personal AI router, discovers compatible computers on a local network and distributes AI workloads across them. Its release adds RTX Spark support, llama.cpp support, direct model downloads from Hugging Face and customizable inference-engine settings.