Perplexity Portable Computer’s arrival on compatible Windows RTX PCs puts a practical question in front of teams: which agent tasks should stay on the machine, and which genuinely need the cloud?

NVIDIA says Perplexity Portable Computer is now available in the Perplexity Windows app for GeForce RTX PCs and RTX PRO workstations with 24GB or more of VRAM. NVIDIA describes a local agent that can work across files and connected tools, with an option to ask permission before escalating harder tasks to cloud models. Availability, hardware requirements and product claims in this article are attributed to NVIDIA.

Local does not mean isolated

The attractive part of a local agent is not merely avoiding a credit meter. It is the ability to inspect sensitive files without sending their contents to a hosted chatbot by default. That can be useful for a finance team sorting statements, an engineering team triaging a local repository, or an operations analyst exploring a customer export.

But a local model is only one boundary. A workflow with Outlook, OneDrive, Slack, GitHub or Google Drive connectors can still reach external services. Teams should evaluate each connector’s permissions, logging and retention rules separately. “Runs on device” is a useful property, not a complete security review.

The split that makes sense

Use local execution when the work is repetitive, bounded by a known file set and sensitive enough that uploading everything would be inconvenient or risky. Examples include extracting fields from documents, clustering a local folder, drafting a review of a pull request, or locating inconsistencies in a spreadsheet.

Use a cloud model when the task needs broader research, a larger context window, shared collaboration or reasoning that exceeds the local setup. The valuable interaction design is the handoff: tell the person what must leave the device, why it is needed and let them approve that step.

Start with a small, auditable workflow

  1. Choose one folder or repository. Do not connect every account on day one.
  2. Define the output. For example, “list stale documentation with the file and supporting line,” rather than “clean up our docs.”
  3. Review actions before they run. A local agent can draft an update; a person should decide whether it is sent, merged or deleted.
  4. Record the boundary. Note which tasks stayed local and which required cloud escalation. That gives the team a concrete basis for future policy.

This is an exciting direction for personal AI. Strong local hardware can turn a private collection of files into something an agent can actually work with, without forcing every first step through a remote service. The best use cases will be the unglamorous ones: tasks where keeping context close to the device makes the work easier to trust.

What to watch

NVIDIA lists 24GB of VRAM as the threshold for the Windows release, which limits the first audience. The next useful evidence is real-world performance across ordinary business files, connector permission behavior and the quality of the consent prompt before cloud escalation. Those details will determine whether local agents become a default workflow or a specialist tool.

Sources and update record

Last updated September 19, 2026. Hardware support and product access may change; this page will be revised when Perplexity publishes additional Windows technical documentation.