Perplexity and Nvidia Launch ‘Portable Computer’ , A Local AI Agent With Zero Token Costs
Perplexity and Nvidia are bringing more AI work out of the cloud with Portable Computer, a local-first version of Perplexity’s Computer agent. Despite its name, the product is software designed to run on users’ own hardware, starting with Nvidia DGX Spark and Linux computers equipped with compatible Nvidia RTX GPUs.
The full agent stack, including models, planning tools, connectors, and security sandboxing, runs locally. That means users can ask the system to analyze files, manage documents, work with code, or coordinate multi-step tasks without automatically sending their data to a remote server.
The biggest headline is cost: work completed by local models carries no per-token inference fee and does not consume Perplexity credits. Cloud models remain available for more difficult assignments, but the system is designed to request permission before escalating a task. In other words, “zero token costs” applies to local work: not optional cloud processing.
According to VentureBeat, the initial PC requirement is demanding: Linux users need an Nvidia RTX GPU with at least 24GB of video memory. Nvidia’s official announcement and local AI coverage position DGX Spark as the first optimized platform, with Windows support planned.
For everyday users, the appeal is straightforward: private files can stay on the device, while frequent AI workflows avoid metered cloud usage. For developers and small organizations, it could make long-running agents more practical: provided they already have powerful hardware.
Before upgrading, check GPU memory, operating-system support, subscription requirements, and whether your work truly needs a local agent. Would you trust an AI assistant more if your files stayed on your own computer? Share your perspective with the Brownstone Worldwide community.
