SANTA CLARA, October 14 NVIDIA unveiled RTX Spark, a new PC platform that merges gaming, creative studio, and personal AI supercomputer capabilities into a single superchip with up to 128GB of unified memory and 1 petaflop of AI compute. The announcement came last week alongside Microsoft, setting the stage for laptops to become available starting Friday, October 16. Ahead of the launch, NVIDIA and its ecosystem partners are rolling out software features and performance optimizations to expand local AI on RTX Spark.
Agent popularity has surged this year, with projects like OpenClaw and Hermes topping GitHub popularity charts. Local large language models have surpassed an intelligence threshold, with models like Qwen 3.8-Flash-Next matching frontier cloud models from a few months ago, resulting in local LLM downloads being five times higher than all of last year.
NVIDIA is partnering with the local AI ecosystem to make RTX Spark ideal for running local agents, featuring a ‘Getting Started’ screen at first boot to choose and install agent apps. These agents use new secure sandboxing technology via NVIDIA OpenShell and MXC, where MXC provides OS-level security and OpenShell simplifies MXC adoption with capabilities like credential protection and security monitoring. Hermes Agent will integrate granular OpenShell controls and MXC by the end of October, while OpenClaw and GitHub Copilot already support MXC.
After installing an agent, setting up local AI is seamless in GitHub Copilot, Hermes Agent, OpenClaw, Perplexity, Pi, and WorkBuddy, which recommend optimized local models and configurations for RTX Spark and set them up in a few clicks. For users wanting both local and cloud AI, NVIDIA Switchyard — an open-source model router library — enables agents to intelligently route queries between local and cloud LLMs to minimize cloud usage while maintaining results comparable to cloud-only execution.
Hermes will add Switchyard later in October, and agents like Perplexity Computer and GitHub Copilot have integrated their own routing solutions. Software providers have simplified agent connections to top Windows apps using easy-to-use MCP libraries and CLIs, with NVIDIA providing MCPs for its applications including NVIDIA App, Broadcast, PAIR, Omniverse, RTX Remix, and a large library of CUDA-X Skills.
NVIDIA is releasing AI playbooks for RTX Spark on Friday, October 16, guiding users to set up their first agent, build an AI chief of staff, configure systems for development or design, or create a creative assistant. NVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB, a new LLM purpose-built for local agents on RTX Spark, will be available Thursday, October 15. This 74B parameter, 9B-active-parameter mixture-of-experts model is packed into 24GB using 2-bit quantization with advanced quantization-aware training to retain quality, aiming to balance speed and intelligence.
Users can run Nemotron 3.5 Bolt 74B at launch via llama.cpp, vLLM, Ollama, Unsloth, and LM Studio, and directly through agents like Hermes and OpenClaw. Like other Nemotron models, Bolt 74B is fully open source — including training data, weights, and recipes — allowing developers to fine-tune and customize agents.
Local AI can extend beyond a single PC via NVIDIA PAIR, a personal AI router officially moving out of beta today, which simplifies tapping into total AI compute across compatible PCs on a local network. The free, open-source tool automatically discovers and connects devices, installs AI models and inference engines, and distributes workloads. The official release adds llama.cpp support, direct Hugging Face model downloads, customizable inference engine settings, RTX Spark support, and various inference and user experience improvements.
PAIR works on Windows, Linux, and macOS.
