Perplexity Computer

Perplexity Computer is Perplexity's general-purpose agent, which the company describes as a digital worker. It coordinates several AI models to run multi-step projects end to end, such as research and report building, filling out forms, shopping and long-running monitoring, using a real browser, filesystem and tool integrations in an isolated environment. Perplexity launched it on Feb 25, 2026 for Perplexity Max subscribers, a plan MacRumors puts at $200 a month.

Two later releases moved it closer to the user. Personal Computer for Mac rolled out on Apr 16, 2026 and combines a Mac with Perplexity's server environments, and since Oct 30, 2026 a Chrome extension lets the Mac and Windows apps work inside the user's own signed-in tabs in Chrome, Edge, Brave and Arc.

Shopping agents from Perplexity have also been to court. Amazon sued over Perplexity's Comet browser agent shopping on Amazon, won a preliminary injunction in March 2026, and saw the Ninth Circuit vacate it on Aug 4, 2026, holding that the human user, not the agent's developer, accesses the site. Perplexity has not published usage numbers for Computer, and Verisoul does not name it yet, so we have no volumes to report.

How Perplexity Computer shows up on your site

Each Computer task runs in an isolated environment on Perplexity's servers with a real browser and filesystem, and it uses the same interfaces a person does. Perplexity's crawler documentation covers only two agents: PerplexityBot, its crawler, and Perplexity-User, the fetcher that grabs a page when someone asks Perplexity about it. Neither describes Computer's browser.

In extension and Mac modes, Computer works in the user's real browser on their home network, so it carries their cookies, their IP and their device.

Why Perplexity Computer is hard to detect

There is nothing to match. Perplexity-User declares itself and comes from a published IP file, which makes Perplexity's simple page fetches easy to name, while Computer's cloud browser has no documented token, signature or IP list. A rule written for Perplexity-User will not catch it.

Perplexity's track record is another reason not to wait for it to declare itself. On Aug 4, 2025, Cloudflare reported that when Perplexity's declared crawlers were blocked, an undeclared crawler posing as Chrome on macOS fetched the same pages from unlisted IPs that rotated across networks, about 3 to 6 million requests a day next to 20 to 25 million from its declared agents, and removed Perplexity from its verified bots. Perplexity disputed the finding and said the traffic came from BrowserBase, a hosted browser service it uses. That story concerns Perplexity's fetchers, not Computer, but it shows why a rule that waits for a Perplexity token can miss Perplexity traffic.

The local modes make it harder still, because IP and user agent checks see a normal user in a normal browser. What remains is the session itself. Academic work on AI browsing agents, studied on Perplexity's Comet rather than Computer, found tells such as text pasted programmatically instead of typed, pointer jumps straight to click targets and short bursts of scrolling, and those are the kinds of signals that still work when the network and user agent look human.

How Verisoul detects Perplexity Computer

Verisoul does not yet name Perplexity Computer, and we would rather say so than guess. Its sessions still get caught: cloud browser and automation markers in the device count toward the bot score, as do automated clicks, typing and navigation. Perplexity's separate page fetches declare Perplexity-User, and Verisoul names those perplexity_user. A Computer session today looks like this in the API.

"agent": null

• bot_score is the probability the session is automated, here 0.97, which is the field to write rules against for Computer
• agent is null, meaning Verisoul flagged automation but has not attributed it to a named agent
• When a session is named, the agent object carries id, type, detection_method and detection_signals, as described in the docs

The Agent Atlas lists every agent Verisoul names today, including Perplexity-User, and the AI agent detection docs explain how bot score and named agents fit together.

How to detect Perplexity Computer step by step

1. Install Verisoul on sign-up, login, checkout and payout pages so every session returns a bot score
2. Write bot score thresholds per step, since Computer has no agent id yet, using the AI agent detection docs
3. Handle perplexity_user page fetches with their own rule, separate from Computer sessions
4. Send high-scoring sessions on sign-up, checkout and payouts to review and watch for a named id as Verisoul adds one

Should you block Perplexity Computer?

Computer usually works for a paying Perplexity Max subscriber, so most teams let high-scoring sessions browse rather than block them on sight. Without a name or a signature you can't tell Computer apart from other automation, so the decision has to rest on what the session is trying to do.

Send high bot scores to review on sign-up, checkout and payouts, where an unnamed agent with a real browser can do the most damage. That rule catches Computer along with every other undeclared automation, which is the protection you want while agents like it stay anonymous.