Perplexity launches Computer for Enterprise with sandboxed agent work and admin controls

Perplexity introduced Computer for Enterprise on March 12, 2026 as a new way to run AI-driven research and execution workflows inside a governed enterprise environment. The product is positioned less like a chat assistant and more like a managed runtime where teams can investigate, browse, automate, and run code without stepping outside the company’s security perimeter.

According to Perplexity’s official description, Computer for Enterprise is built on top of the company’s existing enterprise platform rather than being a separate consumer feature set. That means it inherits SOC 2 Type II, SAML SSO, audit logs, and administrative controls that enterprise customers already expect. Perplexity also says every task runs in an isolated cloud environment, with authenticated integrations plus secure browser and code execution inside the sandbox.

• Wide Research and parallel web search for questions that require investigating multiple items at once
• Scheduling and automation for recurring tasks
• Enterprise governance features inherited from the broader Perplexity platform
• Sandboxed browser activity and code execution for safer tool-using workflows

The launch is notable because enterprise buyers are moving beyond answer generation and toward systems that can complete multi-step work. That raises a different set of requirements: identity controls, auditability, administrative policy, secure execution, and predictable isolation between tasks. Perplexity is explicitly trying to meet that bar by packaging execution around governance rather than treating governance as an add-on.

The broader product context matters too. Around the same rollout, Perplexity also expanded Comet Enterprise and new developer APIs, giving the company a fuller stack that spans discovery, browsing, orchestration, and execution. In that setup, Computer for Enterprise becomes the operational layer for getting real work done, while search and browser products feed context and inputs into the workflow.

That reflects a larger shift in enterprise AI. The competition is no longer only about which model answers best in isolation. It is increasingly about which vendor can provide a runtime that organizations are actually willing to deploy into regulated and audited workflows. Perplexity’s pitch is that agentic work becomes more useful once it is bounded by enterprise controls, not in spite of them.