Perplexity brings its local AI to PCs with AMD Ryzen
Perplexity is bringing Portable Computer to PCs with AMD Ryzen AI Max and Ryzen AI Halo to run AI tasks locally on Windows. The system analyzes files and automates workflows without using Computer credits, but it can ask for permission to use the cloud when it needs current data or advanced reasoning.

Perplexity is expanding Portable Computer so that certain PCs with AMD Ryzen AI Max processors can run AI tasks directly on the device. The feature is available to individual and business Pro and Max subscribers through the Perplexity app for Windows.
The idea is simple: your files can stay on your computer while an agent analyzes them, organizes information and handles repetitive tasks. Local inference does not use Computer credits, although the system can turn to the cloud when it needs up-to-date information or more advanced reasoning.
What Portable Computer can do
Portable brings together several components you would normally have to configure separately: a local model, tools for working with files and applications, a task planner, an isolated environment for running code and a system that detects sensitive content.
You can give it access to specific folders and connect services such as:
- Gmail and Outlook
- Slack
- GitHub
- Compatible applications through local MCP servers, a standard for connecting AI models with tools and programs
In practice, a finance team could schedule a daily review of invoices against contracts and receive a list of discrepancies. An operations manager could get a summary of emails and Slack messages every morning, along with draft responses for pending issues.
It can also review contracts for missing or unusual clauses and classify code change requests as ready, blocked or in need of review.
Recurring tasks use the same local model each time they run, so they do not add to Computer credit usage. However, they only run while the app is open and the PC remains powered on.
The cloud steps in when needed
Working with local files does not mean giving up access to information from the internet. For example, a financial analyst could use a local model to review a forecast file and then request current market data or a more complex analysis of the effect of interest rates.
Portable can combine files on the device with cited results from Perplexity Search and the reasoning capabilities of more than 15 advanced models available in the cloud. Before sending information off the computer, it asks the user for permission.
That step matters. The system does not decide on its own that a sensitive document can leave the device.
Limits for files and sensitive data
Portable runs code and tools inside an isolated environment on the computer itself. Read and write access is limited to the folders the user authorizes, and connected applications are available only through the configured permissions.
In addition, a local classifier reviews content before it leaves the PC. It can detect names, account numbers, credentials and other sensitive data. If it finds any, you can:
- Authorize sending it to the cloud
- Keep that part of the task on the local model
- Stop the process
For accounts managed by an organization, the administrator can also decide whether employees are allowed to use local inference.
Requirements and availability
The feature is available on Windows 10 and Windows 11 for computers with one of these components:
- AMD Ryzen AI Halo development platform
- AMD Ryzen AI Max processor with at least 24 GB of GPU-accessible memory
Initial setup requires approximately 20 GB of free space. Available models include Qwen 3.8 27B and PPLX 27B, the model fine-tuned by Perplexity.
To get started, install the Perplexity app for Windows, open Settings, go to Local Inference and download a model. Then, under Permissions, enable file system access and choose the permitted folders. When creating a task, select a local model from the relevant menu.
For you, the main change is that some routine tasks can run with more privacy without using credits each time. The next thing to watch is how the system behaves when a task combines confidential data with web searches: asking for permission before uploading information helps, but the final decision will still depend on how well you configure folders, connections and authorizations.