In-depth analysis · About 5 min read
Perplexity Computer goes local on Mac: models, cloud escalation and access conditions
Perplexity describes local planning and execution on Apple silicon, alongside an optional hybrid route. We examine model choices, subscription access and the limits of its claims.
- Published here
- Source published
- Source checked
Perplexity’s October 20 announcement describes a Mac mode that runs the model, agent harness and task execution on the device. This analysis records that earlier release rather than treating it as a new announcement on our publication date. The full article explains execution choices and access conditions beyond the local-mode label.
Listen to the news
Read this article aloud or keep listening to the news.
Preparing the reader…
Device voice
Speech is handled by your browser or device. AISubHub does not send the text to an audio API. Voice availability and playback with a locked screen vary by device.
Resuming may repeat a short phrase.
At a glance
- The announced local choices are Qwen 3.6 and PPLX Pro, described as a post-trained Qwen. The article does not identify PPLX Pro’s exact base-model version.
- Local Computer is offered to Pro, Max and Enterprise subscribers on supported Apple silicon Macs. At least 36 GB of memory is recommended; the article does not call that a minimum.
What actually moves onto the Mac
The announcement’s meaningful distinction is the execution stack. It says the model, harness and task execution can all run on the Mac, including planning, reasoning and file processing. That goes beyond using a cloud assistant merely to reach local documents. It also explains why the company presents this as a further step after its earlier Mac integration and Hybrid Compute.
The two named local options should be kept distinct: Qwen 3.6, and PPLX Pro, which Perplexity describes as a post-trained version of Qwen. The article does not specify the latter’s base-version number. We therefore do not silently equate PPLX Pro with another numbered Qwen release or infer capabilities from that name.
Fully local and Hybrid Compute start in different places
Perplexity says Hybrid Compute begins in the cloud, combining research and reasoning there with processing private information on the Mac. Fully local Computer begins on the Mac and keeps the task’s planning, reasoning and file processing on-device. These are two execution arrangements within the Mac app, rather than interchangeable labels for one privacy promise.
The local orchestrator can also seek authorized cloud escalation for current information, browser use and connected apps, with access to 15+ frontier models for advanced reasoning. Our interpretation is that a local starting point does not eliminate a workflow’s possible cloud dependencies. Permission to escalate changes its boundary; an offline claim should be assessed for the specific task and selected mode.
Cross-app work is the proposed use case, not an independent test
The supported-app examples include Notes, Messages, Mail and Numbers, with access subject to user permission. One described workflow combines a to-do list, a teammate’s budget request and spreadsheet figures into a status update and meeting invitation. Its significance is coordinating several local inputs and tools rather than simply generating chat text.
The company also illustrates a portfolio review using private holdings and downloaded market data, a client proposal assembled from confidential notes and an internal rate sheet, and morning priorities drawn from messages and mail. Excel, Keynote and Outlook appear in these examples. They are vendor scenarios, not workflows reproduced by AISubHub, and do not prove every app integration or proposed output works on every machine.
Membership, hardware and setup are part of the offer
The announcement names Perplexity Pro, Max and Enterprise subscribers and supported Apple silicon Macs. Its 36 GB-or-more guidance is a recommendation, not a published minimum. It does not provide a complete supported-device matrix or an exact macOS requirement, so those missing details should not be replaced with assumptions about universal Mac compatibility.
Getting started involves downloading the current Mac app and downloading and setting up a local model through the app. This matters when reading “offline”: task execution after setup is different from an installation that never needs a network. The article does not provide enough pricing or billing detail to call this a free subscription, promise zero charges, or derive a new price comparison.
Permissions and sandbox claims still need verification
Perplexity says code and tool execution use isolated sandboxes with controlled access to files and connected apps. It also says sensitive work can remain on the Mac from planning through output when both model and agent are local. Those are the company’s architectural and security statements; the article is not an independent security assessment.
For readers, the useful distinction is between where a selected task is executed and what all software components might communicate. Neither the examples nor the local-mode label establish that no data ever leaves the device. Performance, network behaviour, access eligibility and privacy guarantees remain untested here; the source supplies no benchmark from which to rank speed, quality or cost.
Why it matters
Our reading: the change expands where an agent can do work, rather than making every Mac-app feature permanently offline. Local execution and authorized cloud assistance are separate choices with different data and network boundaries.
Limits and availability
These are vendor descriptions and illustrative workflows. We did not install the app, inspect its network traffic or audit its sandbox. The article does not establish exact macOS compatibility, benchmark results, detailed plan pricing or guaranteed eligibility for an individual account.
Sources and evidence
Source published: · Source checked:
AI-assisted editorial articles checked against linked primary sources. Vendor claims are attributed; inclusion is not endorsement.