News, Fast

Atom Brief

Aug 28, 2026 · Archive
Tech

Perplexity Ties Its Free Local AI Agent To Nvidia's Hardware

Portable Computer plugs into Gmail, Drive and GitHub on-device, and only reaches Perplexity's cloud servers when a subscriber approves escalation.

  • Requires Nvidia's DGX Spark ($3,999 and up, 128GB unified memory) or a 24GB+ VRAM RTX Linux box.
  • Runs the open Qwen3 27B model and Perplexity's own PPLX 27B directly on the device.
  • Live now for Pro and Max subscribers on Linux; Windows support arrives later.

Why it matters: Local-first here means free of Perplexity's meter, not free of Nvidia's silicon or Perplexity's own servers.

Perplexity AI official blog ↗ · Aug 28, 20268/28/26