Perplexity Computer now runs fully local on Mac: Qwen 3.6 or PPLX Pro, 36 GB recommended

If you have an Apple silicon Mac with 36 GB or more of unified memory and a Perplexity Pro, Max or Enterprise plan, you can now run Perplexity’s Computer agent with no cloud model in the loop: the model, the agent harness and task execution all stay on the Mac. Perplexity announced the mode on October 20, 2026. For anyone weighing local agents, it is one of the first mainstream products to ship an on-device agent loop rather than just an on-device model.

Builder question this answers: what does “fully local” actually keep on the device, and what Mac do you need for it?

What runs where

Perplexity’s Mac app now offers two ways to run a Computer task:

• Fully local. The task starts on the Mac. Per Perplexity, planning, reasoning and file processing happen on device, and with permission Computer works across local files and supported apps including Notes, Messages, Mail and Numbers.
• Hybrid Compute (launched in September). The task starts in the cloud. Cloud models handle research and reasoning, while a local model works with private files and, according to Perplexity’s product page, flags personal information before anything reaches the cloud.

Local mode is not sealed off: you can authorize Computer to escalate to the cloud for current information, browser use and connected apps, and the local orchestrator can call one of 15+ frontier models for harder reasoning. That is when credits get spent, so the local default also gives you control over credit use. Perplexity says code and tool execution run in isolated sandbox environments with controlled access to files and apps, the same protocols used by the cloud version.

Models and hardware

The local mode launches with two model options: Qwen 3.6 and PPLX Pro, which Perplexity describes as a post-trained version of Qwen. Perplexity has not said which Qwen 3.6 size it ships, so don’t assume a specific parameter count.

For fully local runs Perplexity recommends 36 GB of unified memory or more on a supported Apple silicon Mac, citing machines like the new Mac mini with M6 and M5 Pro. MacRumors reports that Hybrid Compute works on Macs with as little as 24 GB, which makes it the fallback for smaller machines.

That 36 GB line is in the same range we see for running capable open models yourself. Our Qwen3.6-35B-A3B local coding guide targets a 32 GB Mac, and our look at laptop memory tiers for local LLMs covers why memory is usually the limit, not compute.

How to turn it on

Perplexity lists three steps:

1. Download and open the latest Perplexity Mac app.
2. Follow the in-app instructions to download and set up a local model (Qwen 3.6 or PPLX Pro).
3. Choose fully local for the task.

What to take from it

If you handle client files, financials or anything you can’t send to a third-party model, local mode lets an agent work on them without the data leaving the machine, while you decide case by case when a task may go to the cloud. If you build your own local agent stack, it is a useful reference point: Perplexity is betting that a Qwen-class model on a 36 GB Mac is good enough to plan and run multi-step tasks across native apps. More local tooling is on our homepage.