Nvidia and Tom's Hardware report that Perplexity's Portable Computer agent is now available in its Windows app, running locally on GeForce RTX and RTX PRO GPUs with at least 24GB of VRAM. It is the same multistep-task agent Perplexity offers in the cloud, but the model, harness, orchestrator and scheduler all run on the machine, built around an on-device Qwen model. Work done locally draws no credits and files never leave the computer. Users can still pick cloud models, and the app asks permission before sending a task off-device. The VRAM floor is the catch — it rules out most laptops and mid-range cards — and a Pro or Max subscription is still required. Local agents are arriving, but for now on enthusiast hardware.