Perplexity AI, the search startup valued at $14 billion, is building a personal computer. Not a metaphorical one. An actual physical box that sits on your desk, runs a custom operating system, and lets AI agents rummage through your local files to get things done. The device costs $50 per month on a subscription basis and represents one of the most aggressive attempts yet by an AI company to insert itself between users and their most private data.
The product, announced this week and reported by Slashdot, is called the Comet. It’s a small desktop machine — a compact slab of hardware roughly the size of a Mac Mini — that pairs with a dedicated monitor and runs an AI-native operating system Perplexity is calling PerplexityOS. The pitch is straightforward: instead of switching between a dozen apps and browser tabs to accomplish tasks, you talk to the computer, and its AI agents handle the rest. Scheduling. Research. File management. Email drafts. All of it mediated by artificial intelligence that has direct access to whatever you store locally.
That last part is what makes this interesting. And potentially alarming.
Most AI assistants today operate in a sandboxed environment. They can search the web, summarize documents you paste into a chat window, or integrate with cloud services through APIs. What they generally cannot do is browse your desktop, open your spreadsheets, read your tax returns, or poke through folders you haven’t explicitly shared. Perplexity’s Comet changes that calculus entirely. By controlling both the hardware and the operating system, the company is positioning its AI agents to operate with the kind of deep local access that has historically been reserved for the user alone — or, less charitably, for malware.
Perplexity CEO Aravind Srinivas has framed the device as the logical next step for AI-powered productivity. In statements accompanying the announcement, he argued that current operating systems — Windows, macOS, Linux — were designed decades ago around paradigms that predate modern AI capabilities. The file-folder-app metaphor, he suggested, is a bottleneck. PerplexityOS is designed from scratch to make the AI agent the primary interface, with traditional GUI elements available but secondary.
The $50 monthly subscription covers the hardware, the operating system, and access to Perplexity’s cloud AI models. There’s no large upfront cost. Think of it as the razor-and-blade model applied to personal computing, except the razor is subsidized and the blade is a recurring charge that compounds over years. For Perplexity, this creates a predictable revenue stream and a stickier relationship with users than its current search product provides. For consumers, it lowers the barrier to entry but creates long-term cost obligations that could easily exceed the price of a traditional PC.
The timing isn’t accidental.
The entire AI industry is racing toward what researchers and executives call “agentic AI” — systems that don’t just answer questions but take actions on behalf of users. OpenAI has been building agent capabilities into ChatGPT. Google’s Gemini is gaining the ability to interact with Android apps. Anthropic released a “computer use” feature for Claude that can control a desktop by reading the screen and clicking buttons. Microsoft’s Copilot is being threaded through every layer of Windows and Office. Each of these efforts, however, is constrained by the fact that the AI company doesn’t control the underlying operating system. They’re guests in someone else’s house.
Perplexity wants to own the house.
By shipping its own hardware and OS, the company sidesteps the permission and integration headaches that plague AI agents running atop third-party platforms. There’s no need to negotiate API access with Apple or beg Microsoft for deeper system hooks. The AI has root-level access by design. Files, applications, network connections, peripherals — everything is accessible to Perplexity’s agents from the moment you power the machine on.
This architectural choice carries significant implications for privacy and security, and the company has already drawn scrutiny on both fronts. Perplexity has a complicated history with trust. In 2024, multiple publishers — including Forbes, Condé Nast, and The New York Times — accused the company of scraping their content without permission to power its AI search answers. Wired published a particularly scathing investigation calling the product a “bullshit machine” that fabricated citations and reproduced copyrighted material. The company has also faced questions about how it handles user query data and whether its privacy policies match its marketing claims.
Against that backdrop, asking users to hand over local file access to Perplexity’s AI agents is a significant ask. The company says data stays on-device by default and that cloud processing is used only when necessary, with user consent. But the details of how that consent mechanism works, what telemetry the OS collects, and whether Perplexity can update its data practices through software updates remain unclear. The privacy policy for PerplexityOS has not yet been published in full.
Security researchers have already raised flags. Any system where an AI agent has broad local file access creates a large attack surface. If a malicious prompt or a compromised web page can trick the agent into executing unintended actions — a class of vulnerability known as prompt injection — the consequences could extend far beyond a bad search result. We’re talking about an AI that can read, write, and potentially delete files on your machine. The stakes are categorically different from a chatbot hallucinating a wrong answer.
Still, there’s a real product vision here that shouldn’t be dismissed.
The frustrations Perplexity is targeting are genuine. Modern knowledge work involves constant context-switching: copying data from one app, pasting it into another, reformatting, cross-referencing, filing. An AI agent with full system access could, in theory, collapse dozens of manual steps into a single instruction. “Find last quarter’s revenue figures in my spreadsheets, compare them with the analyst estimates in my email, and draft a summary memo” — that’s the kind of compound task that current AI tools handle poorly because they lack the connective tissue between applications and data sources. Perplexity is betting that controlling the entire stack is the fastest way to deliver that experience.
The hardware itself is manufactured in partnership with an unnamed Asian ODM (original design manufacturer), according to people familiar with the project. Specifications remain sparse, but early reports suggest an ARM-based processor, 16GB of RAM, and local SSD storage in the range of 256GB to 512GB. The monitor is a standalone display — no built-in compute — and connects via USB-C. The industrial design is minimal, almost deliberately understated, as if the company wants the hardware to disappear and the AI to be the only thing you notice.
Perplexity isn’t the first company to attempt an AI-native hardware product. Humane’s AI Pin launched in 2024 to withering reviews and sluggish sales before the company reportedly explored a sale. The Rabbit R1, a handheld AI device, generated viral preorder interest but disappointed users with limited functionality. Both products suffered from the same fundamental problem: they tried to replace the smartphone without offering enough capability to justify the switch. The Comet is a different bet. It’s not trying to replace your phone. It’s trying to replace your laptop or desktop — or at least become the machine you sit down at when you need to get serious work done.
Whether that bet pays off depends on execution. And on trust.
The subscription model means Perplexity needs to retain users month after month to make the economics work. If the AI agents are buggy, slow, or prone to errors, churn will be punishing. If a privacy scandal erupts — and given the company’s track record with publishers, that’s not a remote possibility — the reputational damage could be existential for a hardware product in a way it isn’t for a free search tool. You can stop using a website in seconds. Returning a computer and migrating your files is a different level of friction.
The competitive response will be swift. Apple has been steadily expanding Siri’s on-device AI capabilities and is expected to deepen local file integration with Apple Intelligence in future macOS releases. Microsoft’s Recall feature — which takes continuous screenshots of user activity for AI-searchable memory — was delayed after a privacy backlash in 2024 but is being retooled for release. Google is building Gemini into ChromeOS. Each of these incumbents has distribution advantages that Perplexity can only dream of. But they’re also constrained by legacy architectures, regulatory exposure, and the need to protect existing business models. Perplexity, as a startup, can move faster and take risks that a $3 trillion company cannot.
Srinivas has said publicly that he views Perplexity not as a search company but as an “answer engine” company, and increasingly, as an AI computing company. The Comet is the physical manifestation of that ambition. It’s a declaration that the future of personal computing isn’t about faster processors or better displays — it’s about an intelligent layer that understands your data, anticipates your needs, and acts on your behalf. Whether users are ready to grant that level of access to a startup with a mixed reputation is the central question.
Early access to the Comet is expected to begin in mid-2025, with broader availability later in the year. Pricing beyond the $50 monthly subscription — including potential tiers for business users — hasn’t been disclosed. The company is reportedly in discussions with enterprise IT departments about managed deployments, though corporate security teams are likely to have pointed questions about data handling and agent permissions.
For now, the Comet exists as a provocation as much as a product. It forces a conversation the industry has been circling for months: how much access should AI agents have to our personal data, who should control that access, and what happens when the company holding the keys has every financial incentive to use what it finds? Perplexity is betting the answer is “more access, controlled by us, and users will be fine with it.”
That’s a big bet. The market will decide if it’s the right one.
WebProNews is an iEntry Publication