Perplexity Computer on Mac: what it does locally, who can get it, and the permissions it asks for

Perplexity Computer on Mac can now run without the cloud doing the thinking. On October 20, 2026, Perplexity said its agent can run "fully local" on supported Apple silicon Macs, keeping "the AI model, harness, and task execution entirely on device." Until then, the Mac app split each task between cloud models and a smaller local model. The new mode is available to Perplexity Pro, Max and Enterprise subscribers, and Perplexity recommends at least 36GB of memory.

This story is based on Perplexity's October 20 announcement, its Hybrid Compute and Personal Computer posts, its help center, setup and pricing pages, and reports by AppleInsider and 9to5Mac, checked October 23, 2026. For how this compares with other agents that operate desktop apps, see GitHub Copilot's computer use.

The short version

• What's new: a fully local mode for Computer in the Perplexity Mac app. The model plans and reasons on the Mac, and the agent works with local files and Mac apps.
• Local models: Qwen 3.6, or PPLX Pro, which Perplexity describes as a post-trained version of Qwen.
• Who: Pro, Max and Enterprise subscribers. Perplexity hasn't named a separate price.
• Hardware: Apple silicon; "36 GB or more is recommended," which Perplexity frames as a recommendation, not a minimum.
• Not local: current web information, browser use and connected apps. Computer can "escalate to the cloud" for those if you authorize it.

What does "fully local" mean?

It means both the model and the agent run on the Mac. Perplexity says planning, reasoning and file processing stay on the device, and that Computer can work with local files and apps "without sending personal data to the cloud." Code and tool actions run in what Perplexity calls "isolated sandbox environments with controlled access to files and connected apps."

That's a change from Hybrid Compute, which Perplexity launched on September 1, 2026. In hybrid mode, cloud models handle planning, web search and reasoning, while a local model processes private files behind a "privacy gate" that can mask details or ask before anything leaves the Mac. Hybrid mode, which needs at least 24GB of unified memory and uses Gemma 4 E4B, Qwen3.6 35B-A3B or a Perplexity model locally, stays in the app; the fully local mode starts on the Mac instead of in the cloud. On Windows and Linux PCs, Perplexity's "Portable Computer" runs the model and agent locally, using PPLX 27B or Qwen 27B depending on the hardware, with 24GB or more of GPU memory.

Last verified: October 23, 2026

You don't bring your own model: Perplexity says you download and set up a local model inside its app. The October 20 post doesn't give the size of either model.

Does Perplexity Computer work offline on a Mac?

Perplexity's announcement doesn't say. 9to5Mac's October 20 report says the local mode includes "running tasks without an internet connection." AppleInsider's report the same day says workflows aren't fully offline, because sending email or using online services still needs a network. Both agree that current information, browser use and connected apps go through the cloud, and that you choose whether to allow that. Perplexity also says "the local orchestrator can call on one of 15+ frontier models for advanced reasoning," a cloud step, without saying whether that needs the same authorization.

Who can get it, and what does it cost?

Local Computer on Mac "is now available to Perplexity Pro, Max, and Enterprise subscribers," Perplexity says. It hasn't announced a separate charge. Perplexity's pricing page lists Pro at $17 a month and Max at $167 a month, both billed annually; it doesn't show monthly-billing or Enterprise prices. Its help center says work handled by a local model "does not consume Computer credits," while any cloud step does. For how these plans compare with ChatGPT, Claude and Gemini subscriptions, see our October 2026 plan comparison.

To use it, Perplexity says to download the latest Mac app, follow the in-app steps to download and set up a local model, and choose the fully local option. The app requires macOS 15 Sequoia or later.

Which Macs: Perplexity says "supported Apple silicon Macs, like the new Mac mini with M6 and M5 Pro," without a full list. AppleInsider points out that the standard M6 Mac mini tops out at 32GB, below the 36GB recommendation.

What permissions does Perplexity Computer ask for on a Mac?

Perplexity's setup and help pages list these macOS permissions for the Mac app:

• Accessibility: "so the agent can click and type," and to identify the windows and apps you're using.
• Screen & System Audio Recording: so it can see what's on screen and capture app previews.
• Full Disk Access: "Search and access files across your Mac."
• Microphone: for voice input.
• Automation: prompted when Computer works with installed apps, "such as Automation for Notes or Mail."
• Folder access: prompts for Desktop, Documents and Downloads when it works with your files. Perplexity's setup page says the agent "only sees what you connect."

Granted permissions are listed under Perplexity > Settings… > Permissions. Perplexity's setup also asks you to install its Comet browser, which handles web tasks, and to pair a phone, which approves sensitive actions and two-factor prompts.

Full Disk Access is the one to think about. Apple said on October 2, 2026 that it will add controls requiring "very explicit user action" before an app gets that permission, because "the risks associated with this level of access will grow substantially" as AI agents become more capable, TechCrunch reported. Apple didn't give a date. An agent that can read every file and click every button also inherits the risks of what it reads: our explainer on prompt injection covers why text inside a file or email can redirect an agent, and our look at agent sandboxes covers what isolation does and doesn't stop.

What's confirmed, reported and uncertain

• Confirmed (Perplexity): the October 20, 2026 launch; on-device model and agent; Qwen 3.6 and PPLX Pro; Pro, Max and Enterprise availability; the 36GB recommendation; opt-in cloud escalation for web, browser use and connected apps; macOS 15 or later; the listed permissions.
• Conflicting: Perplexity's help center still lists 24GB and hybrid-only for Mac; 9to5Mac and AppleInsider differ on how much works offline.
• Uncertain: the full list of supported Macs, the local models' sizes, regional availability, and how the "15+ frontier models" cloud option is authorized.