NVIDIA says its new systems can run local coding agents, test applications, and switch to cloud models when needed.
RTX Spark is designed to bring a developer’s coding-agent workflow onto one computer. NVIDIA says the systems can run local agents, AI frameworks, application tests, and deployment tools without splitting work between a laptop, a cloud model service, and a cloud virtual machine.
The systems use NVIDIA’s GB10 system-on-a-chip and up to 128GB of unified memory. That means the processor and graphics processor share one large pool of working memory, giving local agents more room for code, test output, and conversation history.
Developers can work in Windows or Windows Subsystem for Linux, a Linux environment inside Windows. NVIDIA says Microsoft’s Windows Developer Config and new Arm-compatible software are intended to reduce the need to move code between development and cloud environments.
Supported local-agent options include GitHub Copilot, Pi, Hermes, OpenClaw, and JetBrains’ Junie Local. NVIDIA says Junie Local can read code, edit files, and run tests while keeping the model and code on the machine. Hybrid prompt routers can send requests to local models and escalate them to cloud models when needed.
NVIDIA reports that RTX Spark outperformed Intel’s Panther Lake 12Xe in its own tests: 4.6 times on coding agents, 2.7 times on fine-tuning, and 1.8 times on compilation. The results used specific workloads, so developers still need independent testing across their own projects.
RTX Spark laptops are available now, while small desktop versions are coming in November. NVIDIA has not provided prices or detailed configurations in its announcement, making local-model quality, Arm software compatibility, memory needs, and regional availability the key checks before buying.