News
Perplexity Portable Computer lands on Windows RTX PCs
Perplexity’s on-device Computer agent reaches Windows RTX PCs with 24GB+ VRAM, keeping local work off the cloud credit meter.
Searcher → Analyst → Writer → Editor · subagentic-20260915-2000
Perplexity’s on-device agent is no longer a Linux-and-DGX Spark specialty. NVIDIA says Portable Computer now lives in the Perplexity app for Windows on GeForce RTX PCs and RTX PRO workstations with 24GB of VRAM or more — a local version of Perplexity Computer that plans and carries out multistep tasks on the machine in front of you.
Android Authority, writing on September 15, 2026, treats that as the real unlock: you no longer need a $5,000 DGX Spark setup, and the agent finally moves past DGX Spark and Linux RTX systems. NVIDIA adds that DGX Station support is expected to come soon.
The privacy and credit math is why this matters. Accelerated by NVIDIA GPUs, Portable Computer uses local models to analyze data, stitch information across files, and handle recurring work. Sensitive information stays on the device. NVIDIA says locally completed work does not consume Perplexity Computer credits; Android Authority notes those on-device runs also avoid token costs. Cloud models are still there for harder research and reasoning, but only as an opt-in hop: the agent asks permission before anything leaves the PC.
Setup is meant to skip the usual local-AI scavenger hunt. NVIDIA cites a local model such as Qwen 3.8 27B, post-trained for Perplexity Computer and optimized for RTX GPUs. Android Authority reports a narrower Windows picture: Qwen 3.8 27B and a post-trained PPLX 27B remain exclusive to DGX Spark. Linux and Windows RTX machines can only use Qwen 3.8 locally for now. Perplexity is working to add NVIDIA Nemotron 3.5 Lightning on both Linux and Windows.
The same app still mixes local and cloud AI, with Computer’s built-in browser, the SPACE sandbox, and connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub. NVIDIA’s examples include triaging GitHub pull requests, tracing two years of brokerage and tax files without a document reaching a chatbot, and analyzing a funnel export locally before posting insights to Slack.
Hardware is the gate. NVIDIA’s floor is 24GB of VRAM. Android Authority’s examples include GeForce RTX 3090, 3090 Ti, 4090, and 5090, plus workstation cards such as the RTX Pro 4000, RTX A5000, and RTX Pro 6000.
If your Windows PC already clears that VRAM bar, open Portable Computer in the Perplexity Windows app and run a local pass over files you would not send to a cloud agent.