Perplexity brings Portable Computer local agents to Windows RTX PCs

NVIDIA’s blog says Perplexity’s Portable Computer — a local version of its Computer agent — is now in the Perplexity app for Windows on compatible GeForce RTX PCs and RTX PRO Workstations, using local models for multistep on-device work and asking permission before any cloud handoff.

Local agentic AI just moved from Linux boxes and NVIDIA’s DGX Spark onto the Windows RTX PCs a lot of people already own. The pitch is a PC agent that works on local files without uploading everything — and without burning cloud credits for work that stays on the machine.

NVIDIA’s blog published “Perplexity Portable Computer Is Now Available on Windows, Powered by NVIDIA RTX,” dated September 14, 2026, by Gerardo Delgado. That company blog is the filing event. Portable Computer, as NVIDIA tells it, is a local version of the Perplexity Computer agent — software that plans and carries out multistep tasks. In plain English: a PC agent that works on local files without uploading everything. It is not a new NVIDIA foundation model. This desk did not install the app.

The same post says Portable Computer is now in the Perplexity app for Windows on compatible NVIDIA GeForce RTX PCs and RTX PRO Workstations. NVIDIA says the release builds on existing support for DGX Spark systems and RTX PCs running Linux. File the Windows expansion and the older Linux / Spark support as NVIDIA’s. This desk is not saying every RTX laptop can run it.

NVIDIA says local models analyze data, pull information across files, and handle recurring work on the device. Sensitive information stays on the device, the blog says, and locally completed work does not consume Perplexity Computer credits — the cloud-usage tokens Perplexity charges for its Computer agent. File the local-credit line as NVIDIA’s product claim.

For harder research and reasoning, NVIDIA says Portable Computer can spot when a job needs the cloud, then ask the user for permission before sending anything off the machine. File the permission step as NVIDIA’s. This desk is not saying cloud handoff is off by default for every task.

The app, NVIDIA says, ships with a local model such as Qwen 3.8 27B — a 27-billion-parameter open model — that is post-trained for Perplexity Computer and optimized for NVIDIA RTX graphics chips. Those are company-stated model claims. NVIDIA says users can put the agent to work without researching models or setting up the usual local-AI software stack. Install pins stay in Sources.

Connectors named on the same page: Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub. Tools named on the same page: a built-in browser and SPACE, which NVIDIA calls a proprietary sandbox — a boxed-off workspace on the PC. File the lists as NVIDIA’s. This desk did not hook up those accounts.

Hardware cutoff on the same page: GeForce RTX and RTX PRO GPUs with 24GB or more of VRAM — video memory on the graphics card. NVIDIA DGX Station support is “expected to come soon.” File the 24GB line as NVIDIA’s spec and the Station line as a company plan, not a ship date this desk can name.

CONFIRMED here: NVIDIA’s September 14, 2026 blog — PRIMARY company source. NOT claimed: download counts, revenue, that this desk tested the agent, that NVIDIA trained a new foundation model, that every RTX PC qualifies, or investment advice. Distinct from the already-filed openai-perplexity-gpt6-astra-systems and gemini-app-windows.