Local first, cloud when needed
Perplexity Portable Computer brings its multi-step agent workflow to compatible Windows machines with NVIDIA RTX graphics. NVIDIA’s September 14 announcement describes local models working across files and carrying out tasks on the device.
The key distinction is where a task is completed. Locally completed work does not consume Perplexity Computer credits. More demanding work can be sent to cloud models, with the app asking permission before information leaves the device.
A useful first task
For a first comparison, choose a small set of non-sensitive files and ask a concrete question that can be checked against them. A list of unresolved items across a project folder is easier to evaluate than a broad request to organize your whole digital life.
Check the answer against the files, then inspect whether the task stayed local or used cloud support. The value of a local agent is not just that it runs on your desk. It is that you can understand which work happened there and what information was involved.
The hardware check comes first
This launch is not support for every Windows laptop. NVIDIA specifies GeForce RTX or RTX PRO graphics with at least 24GB of video memory. Ordinary system RAM is a different specification, so check the GPU rather than the headline memory number on the computer.
The announcement is the source for the capabilities and requirements here. We have not benchmarked the app. The attached NVIDIA partnership image identifies the announcement; it is not a screenshot of a completed local workflow.
Source published 2026-09-14. Coverage is based on the maker’s announcement and demonstration.