Skip to main content

Perplexity's Hybrid Compute: Keep Sensitive Data on Your Mac

Perplexity is rolling out a new feature called Hybrid Compute that gives Mac users more control over their data. Instead of sending every request to the cloud, you can now split tasks between cutting-edge cloud models and local large language models running right on your computer. Sensitive information stays on your device, never leaving your hands.

How It Works

When you upload a file or type a message in the Perplexity Mac app, the system automatically checks whether it contains sensitive content. If it does, you'll be asked if you're comfortable sharing that data with the cloud. This is made possible by a new privacy classifier that Perplexity trained specifically for this purpose. It suggests which files and information should stay local, so you can review and confirm before any task is assigned to a model.

Currently, you can choose from a few local models: Gemma E4B and two versions of Qwen 3.6, both with 35 billion parameters. One of those Qwen versions has been post-trained by Perplexity itself. More local models are expected to be added over time. The installation process is seamless—no need to open the Mac terminal. The Perplexity app handles everything automatically. During operation, you can see real-time usage of your CPU, GPU, and memory, and the sidebar shows how many tokens each task consumes. Best of all, tokens generated by local models don't cost anything. You can even queue tasks from your iPhone.

Cloud vs. Local: A Trade-Off

When asked whether Hybrid Compute matches the performance of pure cloud solutions, Jon Staff, Product Manager for Perplexity Mac, was candid: "In short, if we look only at the quality of the original output, the output from using the cutting-edge model alone is almost always better. It is more expensive, but its capabilities are stronger." However, he pointed out that not everyone needs the most powerful model for every task. For many users, data privacy and cost might outweigh raw capability.

Staff described Hybrid Compute as a "tunable scale." Perplexity will try to recommend the most suitable solution based on the context, but ultimately, the user decides what fits their needs best.

Availability and Requirements

Hybrid Compute is currently available only for Mac computers with Apple Silicon chips running macOS 15. Perplexity recommends at least 32GB of unified memory for a smooth experience. The feature is included for Pro and Max subscribers, as well as enterprise customers.

Image

Key Points

  • Hybrid Compute lets Mac users split tasks between cloud and local models.
  • Privacy first: Sensitive data is automatically detected and kept on-device.
  • Local models: Gemma E4B and two Qwen 3.6 variants are available now.
  • Cost savings: Local token generation is free.
  • Trade-off: Cloud models generally produce better output, but local models offer privacy and lower costs.
  • Requirements: Apple Silicon Macs with macOS 15 and at least 32GB RAM.
  • Availability: Pro, Max, and enterprise users.