Perplexity has launched Perplexity Computer, an AI agent system designed to carry out entire tasks rather than answer individual questions. Announced in late February 2026, the product is available to subscribers of the company’s Max and Enterprise Max plans.
The idea behind Computer is that a user states a broad goal and the system plans and executes the steps needed to reach it. In Perplexity’s examples, a user might ask it to research a topic, build a presentation from the findings and email the result; Computer then chains web search, data analysis, document generation and connected services to complete the whole workflow.
Many models, many sub-agents
According to Perplexity, Computer orchestrates 19 different AI models, choosing the one best suited to each part of a job rather than relying on a single model throughout. It also spins up sub-agents that work in parallel: running multiple searches at once, organizing what they find and then producing output files such as slide decks, spreadsheets, reports or simple web applications.
Tasks can run for extended periods in the background, with the finished result delivered when the work is done. That long-running design distinguishes Computer from a chat interface, where the user waits for each reply, and puts it in the same category as agent products from OpenAI, Anthropic and Google that aim to act as autonomous digital workers.
A platform play for Perplexity
Perplexity built its reputation on AI-powered search, and Computer extends the company into general-purpose agents, a market where it competes with far larger rivals. Because it draws on models from several providers, Perplexity is positioning itself as a layer that sits above the model makers, routing work to whichever system performs best for a given step.
The company says Computer can handle business research, market analysis, report writing, presentation building and basic app development from a single instruction. Those are Perplexity’s own claims; independent assessments of how reliably the system completes complex multi-step tasks were not available at launch, and long-running agents across the industry still make errors that require human review.
Access is limited to the company’s highest-priced tiers for now, reflecting the compute cost of running many models and sub-agents per task. Perplexity has not said when or whether Computer will come to its lower-priced Pro plan.
