Perplexity has introduced a new cloud-based AI system called Computer that assigns complex tasks to multiple AI agents operating in different models by breaking them down into smaller parts.

This feature is currently available for Perplexity Max subscribers. According to the company's statement, Computer has the ability to create and execute complete workflows, including tasks that can run for extended periods.

How Perplexity Computer Works

Instead of relying on a single large language model, Computer distributes subtasks among multiple models based on their strengths.

Perplexity states that the core reasoning engine operates on Anthropic’s Claude Opus 4.6. Other models take on more specific tasks:

  • Gemini for deep research
  • Nano Banana 2 for visual production
  • Veo 3.1 for video production
  • Grok for light and speed-focused tasks
  • ChatGPT 5.2 for long context memory and broad searches

Users define the desired outcome (e.g., planning a digital marketing campaign or creating an Android application), and Computer breaks the request into small steps. Each step is delegated to the most suitable model for that part of the workflow.

The entire process takes place in the cloud. Perplexity states that each task operates in an isolated computing environment and has access to a browser, file system, and integrated tools.

Perplexity Tries to Monetize the Emerging Agency AI Interest

This launch comes with increasing interest in "agency AI" systems that can operate more autonomously and manage multi-step processes.

An example of this is Google's tests with Gemini Agencies on Pixel phones.

Previous experimental tools in this area typically operated directly on users' local machines and provided extensive file access and plugin integrations. Despite their power, these approaches raised concerns about reliability, security, and injection risks.

Computer takes a more controlled approach. It operates entirely in the cloud and uses carefully selected integrations. The goal is to provide robust workflow automation while limiting direct access to the user's local environment.

What Perplexity Aims to Achieve with Computer

Perplexity positions Computer as a way to simplify workflows that advanced users have already been manually assembling by switching between multiple AI tools. By coordinating models in the background, Computer aims to make these types of workflows more accessible to users without technical knowledge.

Risks of Computer and Dependency on Large Language Models

Despite its structural design, Computer is still dependent on large language models, which can produce incorrect or incomplete results. Errors can be particularly problematic when tasks involve sensitive data or automated file processing.

As with other AI workflow systems, users will need to carefully review outputs; especially for business, legal, or technical use cases.

Currently, Computer is limited to Perplexity Max subscribers, and the company has not yet announced a wider rollout.