Perplexity introduces Computer, its AI digital worker

Perplexity has introduced Computer, an AI system designed to carry out entire jobs, not just answer questions. The tool breaks a goal into tasks, assigns each one to specialized agents, and can keep the process running for hours or even months, according to the company.

From asking for an answer to assigning a job

The difference from a chatbot lies in the type of task it receives. Instead of asking how to prepare a report, you can ask it to research the topic, gather data, write the document, and deliver it using your connected tools.

Computer breaks that assignment into subtasks. One agent can search the internet while another processes data and a third prepares a document. Coordination happens automatically in the background, so Perplexity says you can leave several jobs running in parallel.

The system can also write code, call connected services, look for additional information, and solve problems that arise during execution. It should only ask for your input when it needs a decision or data it cannot obtain on its own.

Each task runs in an isolated environment with access to a file system, a browser, and various integrations. In practice, that means Computer tries to use software in a way similar to a coworker, rather than limiting itself to generating text inside a chat window.

A system that combines multiple models

Perplexity does not base Computer on a single model. Its proposal is to use the most suitable one for each part of the job, a strategy known as multimodel orchestration: coordinating multiple AI systems according to their capabilities.

At the time of the announcement, Perplexity says Computer uses:

• Opus 4.6 as its main reasoning engine.
• Gemini for deep research and creating subagents.
• Nano Banana to generate images.
• Veo 3.1 to create videos.
• Grok for lightweight tasks where speed matters.
• ChatGPT 5.2 to remember large amounts of context and conduct broad searches.

The selection would not be fixed. Perplexity says it can switch models as more capable ones become available, and that users will also be able to choose which model to use for specific subtasks.

The idea comes from a clear trend: models are not all doing exactly the same thing. Some excel at research, while others are better at images, speed, video, or handling long texts. Computer tries to turn that specialization into a single workflow.

What changes for you

Until now, using multiple models usually meant opening different services, copying results, and deciding which tool worked best for each step. Computer aims to handle that coordination for you.

For example, a company could ask it to analyze a dataset, consult external sources, prepare a report, and create visual materials. The value would not lie only in each response, but in connecting all those actions without requiring you to supervise every transition.

That also introduces important limits. A system that can browse, run code, and use connected services needs clearly defined permissions and oversight, especially when it works for long periods. The announcement describes isolated environments and access controls, but it does not yet provide complete details about its security mechanisms, errors, or usage costs.

Perplexity Computer is already available to Perplexity Max subscribers. The company says it will soon reach Enterprise Max users.

The important signal is that competition is starting to shift from the chatbot to the system that organizes the entire job. What remains to be seen is whether Computer can maintain accuracy during long tasks, control its permissions, and produce reliable results without turning human oversight into another full-time job.