Perplexity Brings Its Local AI Agent to Windows

Perplexity has released its local-first AI agent, Portable Computer, for compatible Windows PCs powered by Nvidia RTX GPUs with at least 24GB of VRAM. The agent runs locally on the device, processing tasks without sending them to cloud models unless permission is granted. This marks a shift from the previously announced September launch date to an available product, answering the main hardware requirement for local inference.

The Windows release is supported on Nvidia GeForce RTX or RTX PRO GPUs with the specified VRAM. Perplexity’s local model, PPLX 27B, is currently available on Windows, while Qwen 3.8 27B is not. Windows users enable local inference in the Perplexity app, which then checks GPU memory and hardware compatibility. The agent can interact with various documents, spreadsheets, PDFs, code, images, and other files, maintaining isolated execution environments for code and tool execution.

It can also connect with approved apps like Gmail, Outlook, Slack, and GitHub through Model Context Protocol servers. Tasks completed by the local model do not consume credits, while those requiring current web information or stronger reasoning call Perplexity Search or cloud models. The local-first architecture requires IT teams to review permissions, data flow, and endpoint policies, as well as verify connector scopes, file permissions, cloud escalation logging, sandbox boundaries, model update controls, and tool execution restrictions for managed users.

Portable Computer is currently available to Pro and Max subscribers across individual and enterprise plans, but broader deployment depends on the organization's endpoint, identity, and data-governance policies.