Perplexity launched Hybrid Compute for its agentic Computer platform on Mac, a setup that splits a single task between frontier models in the cloud and smaller open-weight models running on Apple silicon, The Verge and VentureBeat reported on September 1. Cloud models handle reasoning, web search, and planning. An on-device model takes work that touches sensitive information and private files, The Verge said.
The feature is available through Perplexity’s desktop app for enterprise customers that opt in, plus Pro and Max subscribers, on Apple silicon Macs running macOS 15 or later, VentureBeat reported. Perplexity’s product page for Hybrid Compute on Mac describes the same split: Computer moves a task between cloud models and a local model on the user’s Mac so sensitive data can stay on the device.
The key control is what Perplexity calls a Privacy Gate. VentureBeat said it is a company-trained classifier that runs on the device and scans for personally identifiable information before anything is sent to the cloud. When the gate flags sensitive content, the user chooses whether that portion of the task runs locally or gets shared. Perplexity’s product page shows the same consent flow, with options to process the file on the Mac or upload it anyway.
Jon Staff, who leads Perplexity’s macOS and iOS engineering teams, told VentureBeat the cloud orchestrator breaks a prompt into subtasks and routes them. Sensitive work is delegated to a subagent on the Mac, and those tokens do not go to the cloud. Local inference does not burn Perplexity cloud credits; users pay for power and hardware instead, Staff said.
At launch, users can download any of three local models in the Mac app: Google’s Gemma 4 E4B, Alibaba’s Qwen3.6 35B-A3B, and a Perplexity model, according to Perplexity’s product page. VentureBeat reported the same trio, naming a Perplexity post-trained Qwen3.6 35B as the company’s recommended option. The product page lists a 24 GB unified-memory minimum and 32 GB for best results; VentureBeat said Perplexity recommends at least 32 GB for the better tier of local models. Windows and Linux support will come later, VentureBeat reported.
Hybrid Compute is aimed at work that mixes public research with files people do not want off the machine. In demos described by VentureBeat, a lawyer updated a privileged brief against local case files while a cloud agent pulled public case law, a private-equity associate rebuilt a financial model against confidential projections, and a founder started a marketing analysis on iPhone that later used a studio Mac for local customer data.