Perplexity Computer goes fully local on Mac

If you own a beefed-up Mac, you can now hand over your private financial spreadsheets, chat logs, and daily notes to an AI without sending a single byte to the cloud.

Perplexity just launched a fully local version of its Computer agent for Mac. Available to Pro, Max, and Enterprise subscribers on Apple silicon devices, the feature runs the model, task planner, and execution harness completely on device.

Deep local app integration inside a sandbox

Back in May, Perplexity introduced Personal Computer for Mac to read local files, and followed up last month with a hybrid compute setup that split tasks between cloud and device. This update takes the final leap by keeping the entire workflow local.

Once granted permission, the agent interacts directly across local files and native Mac apps. That includes Apple Notes, Messages, Mail, and Numbers, alongside productivity tools like Excel, Outlook, and Keynote.

In practice, it can scan a budget request in Messages, check your project list in Notes, pull spending numbers from a local spreadsheet, and prepare a status update with a meeting invite ready to go.

All code and tool executions run in isolated sandbox environments with controlled file access, keeping the agent from making unapproved changes to your machine.

Powered by Qwen, but hungry for RAM

Under the hood, local Computer runs either Qwen 3.6 or PPLX Pro, a customized post-trained variant of Qwen.

Running multi-step agents on device is demanding. Perplexity recommends a hefty 36GB of unified memory or more, pointing to machines like the new Mac mini with M6 or laptops powered by M5 Pro.

Users are not locked out of the web, either. The local orchestrator can escalate tasks to Hybrid Compute or tap into over 15 frontier cloud models whenever a workflow requires live browsing or heavier reasoning.

Why it matters

For power users dealing with sensitive data like financial audits, private client proposals, or confidential legal notes, this eliminates the risk of cloud data leakage. If your Mac meets the RAM requirements, it is well worth testing now to cut down cloud credit burn and speed up daily routines.

Across the industry, this marks a clear shift from passive local chatbots to active, multi-step desktop agents. When open-weight models running on consumer hardware can reliably manage native apps, cloud providers may see their query volume cut down significantly.

We think Perplexity made a smart architectural move by offering a flexible bridge between fully offline execution and frontier cloud fallback. The main catch, however, is hardware accessibility. Recommending 36GB of RAM leaves entry-level Mac users behind, meaning this powerful workflow will remain an exclusive luxury for pro workstations for now.