Perplexity is turning the humble Mac mini into something a lot more ambitious: an always‑on AI sidekick that quietly lives on your desk, watches over your files and apps, and does work for you while you get on with your day. It’s called Personal Computer — and it’s Perplexity’s answer to OpenClaw and the broader “agentic OS” race that’s heating up across the AI industry.

Instead of yet another chatbot tab in your browser, Personal Computer is meant to feel more like an operating system for AI tasks. You plug in a Mac mini, leave it running, and Perplexity’s agents gain controlled access to your local files, applications, and ongoing projects. The underlying promise: give the AI an objective, not a prompt, and let it orchestrate the details — securely, visibly, and with a kill switch if it oversteps.

At its Ask 2026 developer conference in San Francisco, Perplexity CEO Aravind Srinivas pitched this as a philosophical shift: “A traditional operating system takes instructions; an AI operating system takes objectives.” The company now has two flavors of agents built around that idea. Personal Computer runs locally on a dedicated Mac, while Perplexity Computer — launched earlier — lives in the cloud as a general‑purpose digital worker that coordinates a whole fleet of underlying models from OpenAI, Anthropic, Google, and others.

On the Mac side, Personal Computer is basically a permanent bridge between your machine and Perplexity’s cloud agent. The Mac mini acts as a kind of home base: it stays online 24/7, holds your files, connects to your local apps, and exposes them to Perplexity Computer and its Comet assistant in a controlled way. That means the agent can rename a folder of images for your portfolio site, resize them for the web, draft the copy, and even push the changes to your CMS — without you manually dragging and dropping files or writing out each step as a prompt.

Crucially for a setup that sounds a bit like “give an AI the keys to your Mac,” Perplexity is leaning hard on the security angle. Every sensitive action is supposed to require explicit user confirmation; the system keeps an audit trail of what the agent did, and there’s a built‑in kill switch if something looks off. Perplexity is also positioning this as safer than OpenClaw-style agents that run more aggressively in the cloud, arguing that a dedicated local box with visible logs gives enterprises and power users more confidence over where their data lives and how it’s used.

For now, access is going to be tightly controlled. The feature will first roll out to Perplexity Max subscribers — the company’s top paid tier — and only on Mac, with a waitlist gating early adopters. That makes sense when you consider how much is at stake: a misconfigured agent that can reach into email, cloud docs, and local storage is a dream for productivity, but also a nightmare if it goes rogue or gets exploited.

Zooming out, a lot of this sits on top of Perplexity Computer, the cloud agent the company quietly framed as “the new operating system of our lives” in CEO posts and technical explainers earlier this year. Computer is not a single monolithic model; it’s an orchestration layer that coordinates 19 specialist models for reasoning, coding, research, and more, running them in parallel and handing off tasks between sub‑agents. Instead of you bouncing between tools, you hand Computer a goal — “audit our marketing site, fix the broken pages, and ship a new landing page” — and it decomposes that into smaller jobs, from scraping data to generating code to writing reports.

Enterprise is where Perplexity clearly wants this to land. The company is already pushing Computer for Enterprise with SOC 2‑certified infrastructure, SSO support, detailed audit logs, and granular connectors into tools like Slack, GitHub, Notion, CRMs, and data warehouses. Workflows can run asynchronously in the background for days or weeks, pinging humans only when something needs approval or when a long‑running task finishes. It’s the “AI employee” narrative — except this employee never sleeps, never forgets, and has direct API access to half of your stack.

Personal Computer on the Mac is the missing local half of that story. In the OpenClaw world, a lot of agent action lives purely in the cloud; Perplexity’s twist is to graft that orchestration layer onto a physical device you own, with local storage and a clear perimeter. The Mac mini is almost the perfect chassis: relatively cheap, quiet, low‑power, and easy to tuck away in an office or server closet, but powerful enough to run a desktop OS and host connectors. You don’t really “use” the Mac in a traditional sense — most interaction happens through Perplexity’s apps and web interfaces — but it becomes the anchor for a persistent personal context that lives beyond any single chat session.

Perplexity is also trying to ride a broader industry wave. Anthropic has been experimenting with Claude as a local coworker that can read and edit files on your machine, while OpenAI partners and open‑source projects have been chasing OpenClaw‑style agents that operate as digital workers. Everyone is essentially poking at the same question: what happens when the AI stops being an app and starts behaving like the computer itself? Perplexity’s bet is that the winner won’t just have the best model, but the best orchestration, memory, and interface — plus a story about security that enterprises and privacy‑conscious users can actually live with.

There are obvious risks and open questions. Cloud‑heavy, multi‑model systems still hallucinate, and stitching together outputs from many models can actually make verification harder, not easier. Compliance and copyright concerns around scraping, automation, and content generation haven’t gone away either, and early adopters will have to think carefully about what they let these agents touch. Cost is another wildcard, especially when long‑running workflows quietly burn through credits in the background; Perplexity’s high‑end tiers are pitched at serious users, not casual tinkerers.

Still, if you squint a bit, you can see why Perplexity is pushing a Mac‑based agent this hard. The company doesn’t own its own frontier model, so it has to differentiate somewhere else — and “AI as infrastructure plus a personal, always‑on computer that feels safer than a black‑box cloud agent” is a pretty compelling answer. For Mac users, especially developers, researchers, and power users who already lean on the mini as a home server or automation hub, Personal Computer looks like the next logical step: turn that box into a full‑time AI coworker that understands your environment, runs your workflows, and lets you stay comfortably human in the loop.