Perplexity launches Hybrid Compute for Mac to enhance AI privacy

Perplexity has launched ‘Hybrid Compute’ for its Mac application, a new feature designed to balance high-level AI reasoning with data privacy. The system functions as a hybrid inference orchestrator that splits tasks between powerful cloud-based frontier models and local large language models (LLMs) running on Apple Silicon.

Under this architecture, complex tasks such as web research and long-horizon planning are handled in the cloud. However, when a task involves sensitive or personally identifiable information (PII), the system delegates those specific steps to a local model on the user’s Mac. To ensure security, Perplexity utilizes an on-device ‘Privacy Gate’—a trained classifier that scans prompts and files to identify sensitive content before any data is transmitted to the cloud.

This approach offers two primary benefits: enhanced privacy for professional users, such as lawyers handling confidential client files, and reduced costs, as tokens processed by local models do not incur the same fees as cloud-based inference. The feature is available for Pro, Max, and enterprise subscribers on macOS 15 or later. Local model options currently include Gemma E4B and variants of Qwen 3.6. Perplexity CEO Aravind Srinivas also announced that the company intends to open-source its PII classifier.