Perplexity Portable Computer Is Now Available on Windows, Powered by NVIDIA RTX
Sep 14, 2026, 8:00 AM · NVIDIA Blog
Perplexity’s agentic Computer now runs locally on Windows RTX PCs with 24GB-plus VRAM, keeping sensitive files on device and only hitting the cloud when you say so.
Why it matters
Local agents stopped being a hobbyist demo the moment a mainstream product shipped them as a first-class mode. Perplexity’s Portable Computer lands in the Windows app for GeForce RTX and RTX PRO machines, building on existing DGX Spark and Linux RTX support.
The pitch is practical: analyze files, stitch information across documents, and handle recurring work with local models so sensitive data never leaves the machine—and local runs don’t burn Perplexity Computer credits. When a task needs heavier cloud reasoning, the agent asks permission before anything goes off-device.
From the desk
We’re watching the privacy story and the platform story at the same time.
On privacy, this is the useful version of local AI. Finance summaries, tax returns, brokerage PDFs—the NVIDIA blog walks through keeping that material on-box while still getting agent help that cites the exact file and page. That’s not a vibe. That’s a reason enterprises and individuals who won’t paste into a chatbot finally have a path.
On platforms, NVIDIA is doing what it does best: make the GPU the assumption. Setup ships with a local model such as Qwen 3.8 27B, post-trained for Perplexity Computer and optimized for RTX. Connectors for Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub pull the agent into the workflows people already live in. Useful AI earns trust when it sits next to work, not when it demands a new religion.
The hard edges are real. You need 24GB or more of VRAM. That’s a premium floor, not a mass-market PC. And hybrid agents that can escalate to the cloud are only as trustworthy as the permission UX—if “ask before sending” gets soft or ignored, the local promise collapses.
I’m for this direction when the evidence holds: keep secrets local, borrow cloud brains only on purpose. The downside if it scales badly is a two-tier world—people with fat GPUs get private agents; everyone else keeps feeding the cloud. Watch whether the VRAM floor drops and whether those permission gates stay sharp.
Context
Portable Computer is framed as a local version of Perplexity Computer that plans and executes multistep tasks. NVIDIA DGX Station support is described as coming soon. The post also plugs other local-model updates on RTX and DGX hardware; those are adjacent product notes, not the core Windows ship.
Who feels it
- Windows power users and professionals
- They get an agent that can touch GitHub, Slack, and Office files without defaulting every document to a remote model—if their GPU qualifies.
- Enterprises with data-residency rules
- On-device completion plus explicit cloud escalation is closer to a compliance-friendly pattern than pure chat-in-the-cloud.
- NVIDIA and Perplexity
- This deepens the RTX-as-local-AI-platform story and gives Perplexity a differentiated Windows surface beyond search.
- Cloud-only agent vendors
- They now face a clearer competitive claim: keep the secrets here, rent the smarts only when needed.
What to watch
- Whether the 24GB VRAM requirement softens as smaller local models improve.
- How often users actually approve cloud escalations—and whether any silent off-device paths appear.
- DGX Station support landing, and whether enterprise IT packages this as a managed local agent.
Companies: NVIDIA