Perplexity Computer

Overview

Perplexity Computer is Perplexity’s agent product for orchestrating models, sub-agents, tools, files, connectors, browsers, and computer environments around a user’s objective.

Current Profile

The first source identifies Perplexity Computer as an agent product associated with reported revenue acceleration. The CEO interview supplies the operating model: different models are assigned to different specialties; sub-agents and tools execute parts of a job; and the system can return completed artifacts such as memos, decks, or research rather than only an answer. Perplexity also describes a future hybrid architecture in which private local data stays on user-controlled hardware while heavier or longer-running work moves to a private server-side environment.

Key Characteristics

• Orchestrates multiple external models rather than relying on one universal model.
• Uses sub-agents, tools, skills, files, command-line operations, and connectors to complete multi-step work.
• Extends from cloud research into browser and computer control through Perplexity and Comet surfaces.
• Supports a proposed local/server split for private context and heavier remote execution.
• Aims to hide API-key, billing, and provider-management complexity from users.
• Requires verification, permissions, recovery, and platform access to make autonomous execution trustworthy.

Evidence

• Product and growth signal: 外卖成为海底捞增长最快的业务，乐高表示不会用 AI 来设计产品 describes Perplexity Computer as an AI agent product and reported revenue-growth driver.
• Multi-model architecture: Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN says the product assigns work across GPT, Claude, Gemini, and other specialized models.
• Artifact and workflow execution: Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN gives company examples involving a board memo, partnership deck, press briefing, files, tool calls, sub-agents, skills, and command-line tools.
• Hybrid execution: Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN describes synchronizing a server-side computer with local hardware for private data and long-running tasks.
• Browser boundary: Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN says many tasks still require opening tabs, filling forms, clicking, and uploading files.

Qualifications

Revenue, valuation, investment, customer, and productivity claims are source-attributed rather than independently verified. Multi-model orchestration can add routing and review complexity, and browser or computer access creates credential, transaction, privacy, and recovery risk. The hybrid personal-computer architecture is described as work in progress, not a demonstrated general solution.

What Changed

• Added the product’s multi-model orchestration, sub-agent, file, tool, and computer-use architecture.
• Added the proposed local/private and server-side execution split.
• Added platform access, permissions, and verification as core product boundaries.

Relationships

• Perplexity - product-owner relationship.
• AI Model Orchestration - specialized model and sub-agent composition relationship.
• Local Agent Execution - local private-data and cloud handoff relationship.
• Computer Use Agent - browser and computer action relationship.
• Platform-Agent Access Conflict - website access and delegated-action relationship.
• AI Application Layer Moat - orchestration and workflow differentiation relationship.
• Nvidia - reported investor and infrastructure ecosystem relationship.