{
  "id": 17794240,
  "title": "My RTX Spark laptop still doesn't have Copilot+ features, so a local LLM on the same machine built its own from one prompt",
  "url": "https://urgent.news/2026/11/02/my-rtx-spark-laptop-still-doesnt-have-copilot-features-so-a-local-llm",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-02T12:30:21.000Z",
  "source": {
    "name": "XDA Developers",
    "slug": "xda-developers",
    "url": "https://www.xda-developers.com/rtx-spark-laptop-doesnt-have-copilot-fatures-local-llm-same-machine/"
  },
  "original_language": "en",
  "account": "Nvidia's RTX Spark laptops feature the company's new NPU, capable of handling 50 or 55 TOPS of AI processing power. However, the RTX Spark machines are still not certified as Copilot+ devices, as features like Recall, Click to Do, and Live Captions are still not available. Despite this, I was able to run Qwen 3.8 Flash Next, a 125-billion-parameter model, on the RTX Spark laptop and have it build its own features using just one prompt.\n\nUsing the RTX Spark recipe on llama.cpp, I tasked the model with creating a system tray app with four features: Recall, Click to Do, Live Captions, and plain-English search. After an unattended run, the model generated a functioning app dubbed Spark Copilot, complete with 60 tests, documentation, and more. This entire process was completed autonomously, with the model writing code, executing tests, and documenting its actions without any external assistance.\n\nThe model had to work within strict constraints, such as not interacting with the llama.cpp server, not accessing the internet, and staying within the system's memory limits. Despite these limitations, it successfully built all four features, which I then tested. The results were impressive - the features worked well, even with some minor quirks. For example, Recall took screenshots every 10 seconds or when the active window changed, while Click to Do allowed me to summarize, explain, and rewrite text or extract tables from screenshots. Live Captions transcribed audio in real time, and file search provided accurate results based on file content and metadata.\n\nWhile there are some limitations, such as the absence of a Windows on Arm build and the need for manual input for certain tasks, Qwen 3.8 Flash Next still managed to create a functional Copilot+ style suite of features on the RTX Spark laptop, demonstrating the potential of local LLMs to replicate and extend the capabilities of Microsoft's Copilot+ ecosystem.",
  "summary": "It called it \"Spark Copilot\", and it works similarly to the real thing.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}