Perplexity has launched Hybrid Compute, a new feature for its Mac-based AI agent platform that splits a task between a powerful cloud-based model and a smaller AI model running locally on the user's computer. The system is designed so that sensitive information stays on the user's own machine while less sensitive parts of a task are handled by a more capable cloud model, addressing both privacy concerns and the cost of using premium cloud AI services.

The feature works by automatically scanning files and information a user tries to upload, using a newly trained privacy classifier to flag content that should stay local, though users can review and adjust these choices before processing begins. Local options include Gemma E4B and two versions of a 35-billion-parameter Qwen model, one of which Perplexity fine-tuned itself, with the company promising more local models to come; installation requires no technical setup, and tokens generated locally are free. Perplexity acknowledges that fully cloud-based outputs remain more capable overall, but argues many users will trade some quality for greater privacy and lower cost. Hybrid Compute is currently limited to Apple Silicon Macs running macOS 15 with at least 32GB of unified memory, and is available to Pro, Max and enterprise subscribers.