Your Mac Can Now Run Perplexity Computer Locally. But Not Every Mac

Perplexity has moved its AI agent a lot closer to your desk. On 20 October 2026, the company said your Mac can now run Perplexity Computer locally, with both the model and the agent doing their work on the machine itself.

It’s available now to Pro, Max and Enterprise subscribers through the Perplexity Mac app. The catch sits in the hardware, and in what “offline” really covers. Here’s the version founders can actually use.

What Perplexity actually shipped

Local mode runs on one of two models: Qwen 3.6, or PPLX Pro, which Perplexity describes as its own post-trained version of Qwen. You download a model inside the app, then pick the fully local option when you start a task.

With your permission, Computer can work across local files and native Mac apps, including Notes, Messages, Mail and Numbers. Perplexity’s own example is a practical one: scan recent messages and emails for deadlines, then build an action list in Notes.

Perplexity also said on X that local tasks don’t use Computer credits. Some coverage points out that the main announcement is less direct on this, so confirm it against your plan before you build a budget around it.

The 36GB catch

Perplexity recommends 36GB or more of unified memory. That rules out a lot of everyday Macs.

AppleInsider notes that on the M5 Pro Mac mini, you’d need the 48GB or 64GB configuration to clear that bar. Perplexity hasn’t published comparable results showing how full Computer workflows perform across different Mac setups. So nobody outside the company can say yet how a 36GB machine feels on a heavy task.

If your team runs on 8GB or 16GB laptops, there’s a lower rung. Hybrid Compute, which arrived in September, splits work between cloud models and a local model. It needs at least 24GB, with 32GB recommended for best results. Perplexity says 8GB and 16GB Macs can’t run it at all.

“Offline” comes with fine print

Several headlines, including the one that likely sent you here, say Computer can now work without internet. 9to5Mac reports that the local mode includes running tasks with no connection. For the model and the agent, that holds.

It doesn’t hold for every workflow. AppleInsider points out that sending an email or reaching an online service still needs a network, even when the AI stays on your Mac. And if a task needs current information, browser use or cloud-connected apps, you can authorize cloud access and decide when to spend credits. At that point the local orchestrator can call on more than 15 frontier models.

So think of it as private by default, online by permission.

There’s one more open question. Perplexity says Computer uses controlled permissions and isolated execution environments, but its announcement doesn’t say whether the local agent needs Full Disk Access. If you handle customer data, ask Perplexity directly before anyone on your team switches this on.

Also Read: Perplexity’s Aravind Srinivas Open-Sources Numbat, a Tool to Catch AI Agents Going Rogue

Why this lands differently for Indian startup teams

TechCrunch has reported that Perplexity treats India as a priority market. Back in February 2024, CEO Aravind Srinivas said the product already had more than a million users here.

Now picture a D2C brand’s operations lead who wants an agent to sort through order exports, supplier invoices and customer emails. Sending all of that to a cloud agent is a call plenty of founders hesitate over. Local mode speaks directly to that hesitation, as long as the hardware is within reach. We haven’t verified India pricing for 36GB-plus Macs, so check configuration costs before you plan around this.

This isn’t a one-off, either. Personal Computer for Mac came in May, Portable Computer for Nvidia’s DGX Spark in August, Hybrid Compute in September, and now fully local mode. Four moves in one year. The direction is clear: Perplexity wants Computer to run wherever your data already lives.

Read More: 25 Indian DeepTech Startups to Watch in 2026: Space, Semiconductors, Robotics and Defence

Our read

For most readers, this is a power-user feature today, not a switch to flip tomorrow. The privacy trade-off is real and welcome. The hardware bar and the unanswered questions on permissions and performance are just as real.

If you’re deciding what to do with it, start here:

• Check your Mac’s memory first. Under 24GB, neither local mode nor Hybrid Compute is an option.
• Test on non-sensitive files before you hand it customer data.
• Ask Perplexity about permissions and confirm how credits work on your plan.

Have you tried it? Tell us in the comments what you’d trust a local AI agent with, and keep an eye on StartupIndiaX for more AI & DeepTech coverage.

FAQs