Perplexity Ships Hybrid Compute to Keep Sensitive Data Local

Perplexity's newest Mac feature routes some of your prompts to models running on your own laptop instead of its cloud. The company calls it Hybrid Compute, and Engadget reports that a built-in privacy classifier scans anything you upload or send and flags what should stay local.

"Any time you try to upload files or send information, we're going to automatically check for sensitive content," Jon Staff, who oversees Perplexity's Mac products, told Engadget. Users review those recommendations before the system splits work between models.

Local runs use Gemma E4B or one of two 35-billion-parameter Qwen variants, one of them post-trained by Perplexity. The app installs the models itself, no terminal required, and shows live CPU, GPU, memory, and token counters while a query runs. Cloud work still goes to frontier models including Opus 5 and GPT-5.6 Sol. Local tokens are not billed.

Staff was blunt about the trade-off. "The short answer is that a fully frontier output is going to almost always be better in terms of raw artifact creation. It's more expensive, it's more capable," he said, framing the choice as "a sliding scale" the user controls.

Hybrid Compute requires Apple Silicon Macs on macOS 15 with at least 32GB of unified memory, and it ships to Pro, Max, and enterprise subscribers, with iPhone task queuing included. That hardware floor sits on the same platform Apple has been leaning on for on-device AI, and it lands in a run of Apple-side stories our tracker has logged, including this week's Mac mini and Studio supply squeeze.

Engadget's write-up leans on the lawyer example: compare privileged case details against public case law without shipping the file to a server. Whether the classifier is accurate enough to trust with that call is a question the piece does not answer.