Perplexity Portable Computer: AI agent now runs locally on Nvidia DGX Spark

With the Portable Computer, Perplexity has introduced a version of its AI agent that runs entirely on its own hardware, without any cloud connection. The software runs on NVIDIA’s new DGX Spark mini-computer and, with some limitations, also on standard RTX graphics cards. For those who process sensitive data or simply no longer want to pay ongoing token fees, this marks the first official Perplexity solution available without a server connection.

What the Portable Computer Actually Does

Until now, every Perplexity agent ran on the company’s servers: requests are sent out, a cloud model performs the calculations, and the response comes back. With the Portable Computer, this process is reversed. The orchestrator, planner, tool router, scheduler, and even the search index run directly on the user’s own device. Only when a task is too complex for the local models—such as during in-depth research—does the system switch to a cloud model like GPT-5.5 or Claude Fable 5 after prompting the user.

Nate, VP of Engineering at Perplexity, describes the goal this way: They’ve brought the same interface you’re familiar with from the cloud version to a fully local app. The comparison to Ollama is obvious, except that, unlike pure model runtimes, Perplexity comes with a ready-made agent system that integrates with Gmail, Google Drive, Slack, GitHub, and Outlook.

Here’s the hardware you’ll need

The reference device is the Nvidia DGX Spark, a compact…