Perplexity brings its local Portable Computer agent to Windows PCs with RTX GPUs

Perplexity brought Portable Computer to Windows on September 14. The local agent had previously been aimed at NVIDIA DGX Spark systems and launched first on Linux, but it is now inside the Windows Perplexity app, letting users run a cloud-free AI agent on an ordinary Windows machine.

The trade-off is an explicit hardware bar: an NVIDIA GeForce RTX or RTX PRO GPU with at least 24GB of VRAM. NVIDIA said support for its DGX Station workstation is expected soon, which keeps the current addressable hardware concentrated among high-end laptops, mobile workstations and desktops with large amounts of video memory.

The bigger difference from cloud agents is where the work happens. Portable Computer runs the model, the agent harness, the orchestrator and the scheduler entirely on the device. Perplexity made the same point when it first shipped the product in August: locally completed work does not consume Perplexity Computer credits, and sensitive information does not have to leave the machine.

The feature is available to Pro and Max subscribers across individual and enterprise plans through the existing Perplexity app for Windows, which users download from the Microsoft Store. After picking a local model from the dropdown, they can pull it down in one click instead of researching models and configuring a local inference stack themselves.

To make the agent useful, Portable Computer ships connectors for Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub, and it can reach other desktop applications through local Model Context Protocol servers running on the Windows device, so the local model can use those tools and data inside automated workflows.

Perplexity offered a logistics example: a controller schedules the agent each morning to reconcile freight invoices against the carrier's local rate sheets, flag duplicate charges or rates that deviate from contract, and generate an exception queue. When a task needs current information or stronger reasoning, the agent can call Perplexity Search or one of more than fifteen frontier models after asking the user's permission.

Why it matters: local agents address two of the biggest pain points of cloud agents at once, since data never leaves the device and repetitive on-device work no longer burns credits. It also shows that competition among agents is shifting from raw cloud model capability toward on-device efficiency and how deeply the agent plugs into everyday office software.

What to watch next: which additional local models enter the picker, when workstation platforms such as DGX Station are formally supported, and whether enterprise buyers lean toward this nearly fully local deployment path to satisfy data-compliance requirements.