Your Mac mini Is Now an Employee: Inside Perplexity's Personal Computer

For most of computing history, the relationship between you and your computer has been simple: you give it instructions, it does them. That relationship is about to change, or at least that's the bet Perplexity is making with its new Personal Computer platform.

Announced at the company's inaugural Ask 2026 developer conference on March 11, Personal Computer is software that runs continuously on a user-supplied Mac mini, giving Perplexity's AI agents always-on access to your local files, apps, and desktop sessions. You tell it your objective, and it works toward it whether you're sitting at your desk, on a flight, or asleep at 3 a.m.

"A traditional operating system takes instructions," Perplexity CEO Aravind Srinivas said at the event. "An AI operating system takes objectives." It's a good line, and it gets at what makes this feel different from the wave of AI assistant announcements that have come before it.

What Personal Computer Actually Does

Personal Computer is not hardware. Perplexity is not building a device. It's a software client that runs locally on your Mac mini while connecting to Perplexity's cloud infrastructure for the heavy AI lifting.

On the local side, the software gives Perplexity's agents direct access to your active desktop sessions, your internal applications, and your files. If you have hundreds of images in a folder that need to be renamed, resized, and prepared for a website, the AI can see that folder, analyze what's in it, and start working. On the cloud side, the actual reasoning and orchestration runs on Perplexity's servers, with the output pushed back to your machine.

The platform connects to the tools most knowledge workers already live in: Gmail, Slack, GitHub, Notion, Salesforce. It can watch for activity across those services and kick off tasks on its own, without you needing to be around. Perplexity describes it as a digital proxy, something that acts on your behalf while you're doing other things.

Users can also choose which frontier AI models power their agent. Claude, Gemini, and Grok are all options at launch, and you can deploy multiple models working in parallel, routing different subtasks to whichever model handles them best. The ability to mix and match models is central to everything Perplexity is building. According to the company's own internal usage data, by the end of 2025 no single AI model accounted for more than 25% of enterprise query volume, down from 90% concentration in just two models at the start of that year.

The Mac mini Moment

It is not an accident that Perplexity landed on the Mac mini. Apple's compact desktop has quietly become the go-to machine for local AI experiments over the past year: cheap, energy-efficient, always-on, happy to sit on a desk and run indefinitely. At $599 for the base model, it's a reasonable way into local AI without buying a server rack.

Perplexity is not the first company to notice this. The concept of using a Mac mini as a local AI host gained mainstream attention after Austrian developer Peter Steinberger built OpenClaw, a personal AI assistant that ran on a continuously powered Mac mini and could be controlled from a smartphone. That project earned Steinberger a job at OpenAI and sparked a wave of similar experiments. Personal Computer is, in many ways, a productized and polished version of that same basic idea, with a major company's infrastructure behind it.

The Privacy Question

Giving any software persistent access to your local files and applications is not a small thing to ask. Perplexity is clearly aware of the trust barrier and has built several safeguards into the product. Every sensitive action requires explicit user approval before it executes. Every action is logged, creating an audit trail you can review. And there is a kill switch, a way to shut the whole system down instantly if something feels wrong.

The company also claims the setup is more secure than comparable systems because of the mandatory confirmation step on sensitive actions. Whether that holds up to scrutiny in practice remains to be seen. But structurally, the AI reasoning happens in the cloud while local access runs through the client on your machine, with the idea being that your most sensitive files never get shipped to a remote server wholesale.

Srinivas was direct about the ambition: "It never sleeps." That's either deeply appealing or quietly alarming depending on your relationship with always-on technology. Probably both.

Who This Is Actually For

Access to Personal Computer at launch requires a Perplexity Max subscription, the company's top tier at $200 per month, which includes 10,000 monthly compute credits. That price point is a clear signal: this is not a consumer product dressed up in enterprise language. It is an enterprise product that happens to run on consumer hardware.

The target user is a knowledge worker or small team that runs complex, multi-step workflows across many tools: the kind of person who currently juggles a dozen browser tabs, manually copies data between applications, and burns hours every week on tasks that are technically automatable but practically annoying to automate. Personal Computer is for them.

Srinivas made the positioning explicit: "Perplexity PC is meant for serious people." He drew a direct contrast with open-source agent setups, which still require significant technical legwork to configure. The pitch, basically, is that you get serious capability without spending a weekend fighting with API keys.

The Bigger Picture

Perplexity is a company in an unusual position. It has no frontier AI models of its own. It sits downstream of the model providers (OpenAI, Anthropic, Google) and its whole pitch is orchestration: the ability to route the right task to the right model at the right moment, automatically, without the user needing to think about it.

That is either a brilliant wedge position or a precarious one, depending on how you read the competitive landscape. The model providers could, in theory, build their own orchestration layers. Microsoft and Google have both been working on exactly that. But Perplexity's argument is that mixing models (using Claude when Claude is best, Gemini when Gemini is best) is something no single provider can offer, which gives an independent orchestration layer a real reason to exist.

Personal Computer is where that bet gets concrete. If it works as advertised, it's not really a tool in the traditional sense. Not a chatbot you query, not an assistant you summon when you need something, but a system that just runs, working through your to-do list whether you're at your desk or not.

The product is currently waitlist-only. Perplexity says early users will get hands-on support as the platform rolls out. For now, the best way to experience it is to sign up and wait, which is, somewhat fittingly, exactly the kind of task you might eventually outsource to it.