Perplexity Launches Fully Local-Operating AI Agent

Perplexity announced "Portable Computer," a local-operating version of its AI agent platform "Computer." The platform targets Nvidia's DGX Spark (a desktop supercomputer) and Linux machines equipped with Nvidia RTX GPUs, and was developed in close collaboration with Nvidia. The defining characteristic is that all processing is completed on the user's own device.

An AI agent is a system that autonomously executes multiple procedures to complete tasks. Currently, many AI agents operate on cloud servers, incurring costs with each use and transmitting files or input content externally. To address these challenges, Portable Computer prioritizes local processing, with tasks initiated first on the device. When sending data to more powerful models in the cloud, the design requires obtaining user permission each time.

For tasks completed entirely locally, no billing credits are consumed. Additionally, since all AI models, user files, and work results can remain on the device, external data leakage risks are minimized even when handling highly sensitive information. During Monday's demonstration, the team performed a document review in a local environment targeting a folder containing multiple tax return documents (1099 forms) and investment-related files. The scenario focused on handling sensitive financial data that users would hesitate to upload to the cloud.

Technically, Portable Computer integrates local models, agent control mechanisms, inference engines, various tools, app integration features, and security sandboxes into a single system. Historically, building an AI stack locally (the suite of software necessary to run AI) required separate steps including model downloads, inference server setup, and tool configuration. Nate, Perplexity's Vice President of Infrastructure and Enterprise, stated in a press briefing that "constructing a local AI stack has been extremely cumbersome," and explained that the company focused on enabling users to begin using Portable Computer immediately.

The announcement also carries strategic significance for Nvidia. Nader, the company's Director of Developer Technology, stated that "local AI has reached a turning point." While enthusiast applications centered on running heavily quantized models have dominated historically, the rise of practical open-source models has changed the landscape. Nvidia has focused on semiconductor sales for large-scale AI data centers in recent years, but this move positions the company as making a serious commitment to the local AI hardware market.

The proliferation of local AI represents a development that could transform how AI agents are typically used, moving beyond the current cloud-dependent paradigm. Particularly in fields such as healthcare, law, and finance—where strict data handling standards apply—the advantages of local processing for information management are likely to be valued. Going forward, the expansion of compatible hardware and integration with more applications and tools will be critical factors determining Portable Computer's adoption.