Perplexity has launched a new tool called “Computer,” designed to let users assign complex tasks and have them completed by a coordinated system of AI agents. The feature is currently available to Perplexity Max subscribers.
The company describes a computer as a system that creates and executes full workflows and is capable of running for extended periods, including hours or even months.
Multi-Agent, Multi-Model Approach
With a computer, users define a desired outcome, such as planning and executing a local digital marketing campaign or building an Android app for research purposes. The system then breaks the objective into subtasks and assigns them to multiple AI agents.
Each agent runs a model selected for its specific strengths. The core reasoning engine relies on Anthropic’s Claude Opus 4.6. Gemini handles deep research tasks, Nano Banana is used for image generation, Veo 3.1 manages video production, Grok supports lightweight tasks where speed matters, and ChatGPT 5.2 is used for long-context recall and wide-ranging search.
This best-model-for-the-task strategy differs from competing offerings such as Claude Cowork, which relies exclusively on Anthropic’s models.
All tasks run in the cloud within isolated compute environments. Perplexity says each task environment includes access to a real filesystem, a real browser, and built-in tool integrations.
The company positions Computer as a way to simplify workflows that advanced users were already constructing manually. Power users often combine multiple AI models and use protocols such as MCP, or Model Context Protocol, to give those models access to local files and applications.
Perplexity Computer takes a different technical approach but aims for a similar outcome: coordinated AI agents operating across user files, services, and applications.
The concept builds on earlier efforts such as OpenClaw, which explored similar multi-agent orchestration. Perplexity’s Computer expands on that idea by packaging the workflow automation into a unified, cloud-based system designed for broader accessibility.
