Perplexity just turned “agents” from a buzzword into a full-blown computing model, and it’s calling that new layer Perplexity Computer. In plain terms, this is Perplexity’s attempt to answer a big question hanging over modern AI: if models are already insanely capable, why does using them still feel like babysitting a very smart intern instead of working with a reliable digital colleague?

Perplexity Computer is pitched as a general-purpose digital worker that doesn’t just answer questions or run a single task, but spins up, coordinates, and completes entire workflows over hours, days, or even months. Instead of you hopping between tools, prompts, tabs, and scripts, Computer is meant to sit on top of your software stack and quietly handle a growing backlog of work in the background, only tapping you on the shoulder when it actually needs input.

The way Perplexity describes it, you don’t start with a prompt so much as a desired outcome: “Launch my new newsletter,” “Clean up and document this codebase,” “Prepare everything for my quarterly business review,” that sort of thing. Computer takes that goal and decomposes it into tasks and subtasks, then spins up sub-agents to tackle each piece. One agent might be crawling the web and internal docs, another is drafting documents, another is transforming data, another is calling APIs – all running asynchronously and in parallel.

Under the hood, though, the real story is orchestration. Perplexity has been loudly model-agnostic for a while, and Computer doubles down on that philosophy. Instead of betting the future of work on one “god model,” Perplexity Computer acts as a conductor for an entire orchestra of models, each playing to its strengths. In its launch configuration, Computer uses Opus 4.6 as the core reasoning engine – the one that decides how to break down work and who should do what – and then dispatches subtasks to a roster of specialists: Gemini for deep research, a dedicated vision model (Nano Banana) for images, Veo 3.1 for video, Grok when speed matters, and ChatGPT 5.2 when you need long-context recall or wide, sprawling searches.

That may sound like marketing boilerplate, but it speaks to a real shift. The current frontier models are diverging rather than converging: one is better at reasoning, another at speed, another at multimodal inputs, another at code, another at search-grounding. The Perplexity thesis is that the “real” computer of this era isn’t a single giant model but a harness that can intelligently route work across many of them, swap them out as they evolve, and even let you choose which model handles which subtask when token budgets or compliance rules become constraints.

From a user’s perspective, Perplexity Computer is positioned as the next step after chatbots and single-task agents. Traditional chat interfaces give you answers. Agent-style tools can perform discrete tasks – book a flight, draft a response, summarize a PDF. Computer is meant to sit one layer up: it creates plans, assembles sub-agents, executes, monitors, and adjusts over time. It lives in a sandboxed environment with access to an actual filesystem, a real browser, and integrations into third-party tools and APIs, which is key: this isn’t just moving tokens around in the abstract; it’s meant to operate the same interfaces you do.

The historical callback in Perplexity’s own announcement is telling. They point to the 18th-century use of the word “computer” – human “computers” who broke down complex astronomical calculations and worked in parallel to predict the path of Halley’s Comet. The modern spin is obvious: the word has changed form, but the core idea remains the same – divide complex work, distribute it across workers, and treat accuracy as non-negotiable. Where early digital computers automated arithmetic, Perplexity Computer is trying to automate the project itself.

Crucially, this isn’t arriving in a vacuum. Perplexity has spent the past couple of years building the pieces you’d expect for such a system: an AI-native browser (Comet) that can reason over the live web, an embedded assistant inside that browser, and a deep research stack designed to ground responses in real sources instead of hallucinated answers. It added persistent memory, so the system can remember what you’ve done and why, and a maturing tasks layer that already flirted with agentic behavior. Computer is essentially the moment all those pieces are snapped together and given a single name.

There’s also a strategic element here: Perplexity has been steadily building a reputation as the “accuracy-first” AI – the one that cites its sources, plays nicely with enterprise data, and doesn’t try to lock you into a single model vendor. With Computer, that same model-agnostic stance becomes a differentiator in the agent wars. Where others wrap a single flagship model in a workflow engine, Perplexity is leaning into being the connective tissue between many flagships at once. If you believe the future belongs to fleets of specialized models rather than one winner-takes-all giant, this is the logical product to build.

Availability-wise, Perplexity isn’t pretending this is a toy. Computer is launching first for Perplexity Max subscribers, with Pro and Enterprise support following after. That’s a clear signal: the company expects early use cases from power users, startups, and teams that already live inside Perplexity for research and want to hand off more of the “doing” to the same environment. There’s even a live stream where Perplexity showcases curated Computer runs in real time – a sort of always-on demo of what a “digital worker” actually does when you point it at real-world tasks.

Of course, a system that runs for “hours or even months” and can quietly operate APIs, browsers, and files on your behalf raises all the usual questions: how controllable is it, how transparent is it, and what happens when it goes off the rails? Perplexity’s framing is that each task runs in an isolated compute environment, with strict tool boundaries and clear checkpoints when it needs human input. They’re selling Computer as a “safe harness” for powerful models – the guardrail layer that enterprises have been demanding before they let AI loose on sensitive workflows.

Zoom out, and Perplexity Computer looks a lot like an answer to a broader industry trend. 2025 and early 2026 were dominated by launches of personal and professional AI agents, from coding copilots to autonomous research bots and experimental platforms that promised “set-and-forget” automation. What most of them lacked was staying power and breadth: they were strong on demos, weaker on long-running, multi-step, multi-tool work that looked like a real job. Computer is Perplexity’s bet that the winners in this space will be the ones that can keep projects alive over time, juggle many different tools and models, and adapt as those models evolve.

The idea of “AI is now the computer” is a neat slogan, but it also captures the shift in how Perplexity wants you to think about its product. Instead of treating AI as an app you open, type into, and close, Perplexity wants it to be the substrate – something you delegate work to, revisit later, and trust to keep operating while you’re doing something else. Whether that vision lands will depend less on today’s glossy launch and more on how reliably Computer behaves once people start asking it to run their real lives and businesses. But as of today, Perplexity has made its move: it doesn’t just want to answer your questions; it wants to be the system that quietly does the work behind the scenes.