Perplexity has entered the agentic AI space with the launch of Computer, a platform designed to accept high-level user objectives and autonomously execute them through a coordinated network of AI agents. The product is currently available exclusively to Perplexity Max subscribers and represents a significant step in the company's ambition to move beyond search. Rather than answering questions, Computer is built to get things done.
According to the company, Computer functions as "a system that creates and executes entire workflows" and is "capable of running for hours or even months." This positions it not as a chatbot or a simple automation tool, but as a long-horizon task executor — one intended to handle complex, multi-step objectives that unfold over extended timeframes. The scope of what it promises is notably broader than most agentic offerings currently on the market.
The user experience centers on outcome-based instructions rather than step-by-step commands. A user might submit a directive such as "plan and execute a local digital marketing campaign for my restaurant" or "build me an Android app that helps me do a specific kind of research for my job." Computer then breaks those objectives into discrete subtasks and assigns each to the most appropriate agent, selecting from a range of underlying models based on what Perplexity determines each task requires.
That model-routing architecture is one of Computer's defining characteristics. Anthropic's Claude Opus 4.6 serves as the core reasoning engine, while Gemini handles deep research workloads. Nano Banana is deployed for image generation, Veo 3.1 for video production, and Grok for lightweight, speed-sensitive tasks. ChatGPT 5.2 is assigned to "long-context recall and wide search." This multi-vendor approach reflects a deliberate philosophy: match the model to the task rather than commit to a single provider's ecosystem.
That philosophy sets Computer apart from competing platforms. Products such as Claude Cowork — Anthropic's own agentic computing environment — operate exclusively within a single model family. Perplexity's willingness to orchestrate across multiple vendors introduces both flexibility and complexity, and it signals a bet that no single model is optimal across all task types.
The infrastructure underpinning Computer is entirely cloud-based with prebuilt integrations. Perplexity states that "every task runs in an isolated compute environment with access to a real filesystem, a real browser, and real tool integrations." This means the system can interact with external services, retrieve and manipulate files, and navigate the web — all without requiring the user to configure local environments or manage technical dependencies.
There is a meaningful precedent for this kind of workflow, even if it previously required considerable technical sophistication to assemble manually. Power users have long constructed their own multi-model pipelines, selecting different AI systems for different tasks and leveraging tools such as MCP (Model Context Protocol) to connect those models to local data and applications. Perplexity Computer is designed to democratize that capability — delivering to a broader professional audience what previously demanded expertise to configure from scratch.
Understanding the context around Computer also means understanding OpenClaw, an agentic tool that serves as a conceptual forerunner to this category of product. Originally released under the name ClawdBot and subsequently rebranded as Moltbot before settling on OpenClaw, the tool was designed to operate as an ambient, background process on a user's local machine. Its capabilities ranged from organizing email archives to constructing websites — an early demonstration that AI agents could function as persistent, autonomous workers rather than on-demand assistants. The trajectory from OpenClaw to platforms like Computer illustrates how rapidly this category has matured and how much infrastructure investment now sits behind what appears to be a simple task prompt.
