Perplexity Computer Now Runs Entirely on Your Mac
Perplexity Computer Now Runs Entirely on Your Mac
The model and the agent both run on-device now — planning, reasoning, and your files never have to touch the cloud. The catch: you need a 36GB Mac and a paid plan.
I've been waiting for an announcement like this. Most "AI agents" are just a fancy chat interface on someone else's server — your emails, your notes, your spreadsheets get shipped off to a data center every time you ask the thing to do something. Perplexity just flipped that arrangement on its head, at least on the Mac.
In a three-post thread on X on October 20, Perplexity announced that its Computer agent can now run fully local on your Mac. Not a local model with a cloud agent directing it — the model and the agent both live on the device. Planning, reasoning, file processing: all of it happens on your machine. As Perplexity put it: "Work with your native Mac apps and local files, and run tasks without sending sensitive data to the cloud."
What actually changed
This is the third step in a sequence Perplexity has been building all year. In May, Personal Computer brought the agent to local files and native Mac apps. In September, Hybrid Compute split the work — cloud models handled research and reasoning while a local model touched your private files. Now the fully-local mode takes the last step: the model, the agent orchestrator, and task execution all run on Apple silicon.
The practical setup, according to TestingCatalog's coverage: install or update the Perplexity Mac app, download and set up a local model through the app, then choose "fully local" when starting a task. The local model options are Qwen 3.6 and PPLX Pro, Perplexity's post-trained version of Qwen — so you're running an open-ish model family, not a black box.
Perplexity's examples of what the local agent can do are concrete and genuinely useful: review project tasks in Notes, pick up requests sitting in Messages or Mail, compare planned vs. actual spending from a spreadsheet, draft a status update, prep a meeting invitation — even portfolio reviews against downloaded market data or morning action lists assembled from local correspondence.
The catches: RAM and a subscription
Here's the part nobody should skip. Perplexity recommends at least 36GB of unified memory for the best local experience, and explicitly cites the new Mac mini with M6 and M5 Pro as reference hardware. That's a heavy bar — most Macs in people's bags right now ship with 16 or 24GB. Local AI that can reason through multi-step work is a memory hog; there's no way around the physics of fitting an agentic workload into RAM alongside everything else you're doing.
The second catch is pricing. Fully-local Computer is available to Pro, Max, and Enterprise subscribers — that's $20/month for Pro and $200/month for Max (US list prices). Free-tier users don't get this. So the headline "your agent, your files, never touches the cloud" comes with an asterisk: if you can afford the RAM and the subscription.
Fully local is a mode you choose per task, not the only way the app works. If a task needs current information, browser work, or connected apps, you can authorize the cloud — where, per TestingCatalog, the orchestrator can call on more than 15 frontier models. Hybrid Compute remains available in the same app. The honest read: Perplexity is betting that some tasks are worth keeping private and some are worth the cloud, and the user gets to draw the line each time.
The numbers at a glance
Why this actually matters for local AI
The significance isn't just a product update. Every major AI agent right now — ChatGPT's agent mode, Claude's, Copilot's — is cloud-native by default. Your sensitive material travels with every task. Perplexity putting a multi-step agent (not just a chatbot) on-device, with isolated sandboxes for code and tool execution, is the first mainstream subscription agent to make "your files never leave the Mac" a real, selectable mode.
The security framing is sensible too: code and tool execution take place in isolated sandboxes with controlled access to files and connected apps, using the same security protocols as the cloud version. Local doesn't mean reckless — the agent still asks before touching things it shouldn't.
A few honest caveats, because Perplexity hasn't disclosed everything: no parameter counts, no context-window size, and no independent benchmarks for the local models have been published. "Fully local" is also the default behavior of the mode you choose — not a hard guarantee that zero bytes ever reach the cloud, since the whole point is that you can approve cloud escalation when a task needs it. Keep those in mind before treating this as an air-gapped system.
Still, the direction of travel is unmistakable. Apple keeps shipping machines with more unified memory, models keep getting more capable at smaller sizes, and now there's a mainstream agent that runs on your desk instead of in a data center. If you've ever wanted to point an AI agent at your most sensitive stuff — tax spreadsheets, confidential client briefs, private correspondence — this is the first big-name option that doesn't require the files to leave the room.
Frequently asked questions
Is Perplexity Computer's local mode free?
No. Fully-local Computer is available to Pro, Max, and Enterprise subscribers only — Pro starts at $20/month and Max at $200/month (US list prices). Free-tier users don't get access to this mode.
Which Macs can run Perplexity Computer fully locally?
You need an Apple silicon Mac, and Perplexity recommends at least 36GB of unified memory for the best experience — it cites the new Mac mini with M6 and M5 Pro as reference hardware. Many existing Macs with 16–24GB won't be comfortable running it.
Can Perplexity Computer work offline?
According to Perplexity's announcement, yes — the model and the agent both run on your Mac, so tasks on local files can work without an internet connection. Tasks that need current information, browser use, or connected apps still require you to authorize cloud access.
Is my data really private in local mode?
That's the design: planning, reasoning, and file processing stay on-device, and files only reach the cloud when you approve it. Code and tool execution run in isolated sandboxes. But it's a selectable mode, not a guarantee — if you authorize cloud escalation for a task, that task's data goes to the cloud.
What's the difference between fully local and Hybrid Compute?
Hybrid Compute, launched in September, splits the work: cloud models do research and reasoning while a local model handles private files. Fully local mode runs everything — model, agent orchestrator, and task execution — on your Mac, with cloud access only when you explicitly authorize it.
The bottom line
Perplexity just shipped the thing privacy-conscious power users have been asking for: a real AI agent that can do multi-step work on your Mac without your files touching the cloud. The hardware bar is high and it's paywalled behind Pro/Max/Enterprise, which will limit who actually uses it. But as the first mainstream agent to make "fully local" a first-class mode, it's a genuine milestone — and a sign of where agentic AI is headed: onto your desk, not into someone else's data center.
If you have a high-RAM Apple silicon Mac and already pay for Pro, it's worth trying this week. For everyone else, this is the roadmap item to watch: the hardware will get cheaper, the models will get smaller, and the "cloud by default" assumption is starting to crack.
Sources and further reading
- 9to5Mac — Perplexity's Computer can now run entirely on your Mac, even offline (Oct 20, 2026)
- TestingCatalog — Perplexity Computer now runs fully locally on Mac (Oct 21, 2026)
- Runtimewire — Perplexity runs its Computer agent fully locally on Apple silicon Macs (Oct 20, 2026)
- CloudZero — Perplexity pricing in 2026: Free vs. Pro vs. Max (Pro $20/mo, Max $200/mo)
