Perplexity launches AI‑agent platform, deploying coordinated teams of bots

Perplexity’s “Computer” arrives just weeks after the open‑source wave sparked by OpenClaw, and it does so with a scale that few rivals can match. According to PCWorld, the platform coordinates up to 19 different AI models in a single workflow, with Anthropic’s Claude Opus 4.6 serving as the core reasoning engine while Google’s Gemini tackles deep‑research tasks, Nano Banana generates images, Veo 3.1 produces videos, Grok handles lightweight operations, and ChatGPT 5.2 provides long‑context recall and broad search capabilities【Pcworld】. The result is a cloud‑native “digital worker” that can be instructed to build a web dashboard, code a simple app, assemble a PowerPoint deck, or even render an animated GIF, delegating each sub‑task to the model best suited for it. Unlike OpenClaw’s locally‑run agents, Computer lives entirely in Perplexity’s walled‑garden infrastructure, interacting with external services through a suite of integrations rather than direct hardware access.

The service is currently gated behind the Perplexity Max subscription, priced at $200 per month, and is marketed as a “general‑purpose digital worker” that operates the same interfaces a human would, according to chief business officer Dmitry Shevelenko’s LinkedIn post cited by PCWorld【Pcworld】. Shevelenko says the tool was born from an “internal experiment” that proved a team’s week‑long effort could be compressed into an overnight run while the developers slept. VentureBeat confirms the same pricing model and adds that the platform’s orchestration layer can spin up sub‑agents on demand, effectively turning a single user prompt into a coordinated multi‑model project pipeline【VentureBeat】.

A key differentiator is how users interact with Computer. Perplexity has chosen to surface the agent through its own web and mobile app, rather than the ubiquitous chat‑based front‑ends used by OpenClaw, Manus AI, or other personal‑assistant projects that live in WhatsApp, Discord or Telegram. Shevelenko explained that the team initially tried a Slack‑style interface because it “felt more like a digital worker than just an agent,” but ultimately rebranded the experience as a “computer” to emphasize its broader capabilities【Pcworld】. This design choice keeps the workflow inside a controlled environment, which PCWorld argues mitigates the risk of rogue agents acting on a user’s local machine, albeit at the cost of sandbox‑imposed limits that OpenClaw’s on‑device model does not face.

The launch also signals Perplexity’s intent to stake a claim in the rapidly expanding “personal AI agent” market that Meta, Google and a host of startups have been courting. VentureBeat notes that the platform’s ability to marshal 19 distinct models gives it a breadth of functionality that rivals such as Meta’s Manus AI lack, positioning Computer as a potential one‑stop shop for enterprises and power users who need end‑to‑end automation without stitching together separate tools. Early adopters are already testing the service for tasks ranging from automated email triage—another Perplexity offering that can draft replies and schedule meetings for $200 a month, as reported by VentureBeat—to more creative pipelines like generating marketing assets on the fly【VentureBeat】.

Analysts have not yet published formal forecasts for Computer’s market impact, but the combination of high‑priced subscription revenue and the ability to replace multiple specialist tools could make it a lucrative add‑on for Perplexity’s existing user base. If the platform lives up to its promise of “massively multi‑model orchestration,” it may force competitors to either open their own cloud‑based orchestration layers or double down on local‑hardware solutions to differentiate on privacy and latency. For now, the real test will be whether users can trust a walled‑garden AI to handle complex, multi‑step projects without the friction of moving data in and out of the sandbox—a question that Perplexity will answer in the weeks and months ahead.