Perplexity has released “Hybrid Compute” for the Mac, a feature that splits a single AI task between powerful cloud models and a compact model running directly on the device – sensitive data is meant to never leave the machine.
How Hybrid Compute works
A request starts in the cloud as usual, where large models handle planning and research. As soon as a step touches personal information, Perplexity says it hands that step to a local model on the Mac without losing the task’s context. The boundary between cloud and device is policed by an open-sourced classifier called PII-Tracer – an encoder model of roughly 0.6 billion parameters with four possible outcomes: keep it local, mask the sensitive spans, refuse the action, or ask the user for consent. Work run locally consumes no cloud credits, according to the company.
Requirements and models
At launch, users can pick from three local models, including Google’s Gemma 4 (E4B) and Alibaba’s Qwen3.6 35B-A3B plus a Perplexity-tuned variant. The feature is available to Pro, Max and Enterprise subscribers and sets clear hardware requirements.
- Apple silicon Mac running macOS 15 or later
- At least 24 GB of unified memory, 32 GB recommended
- One-click setup inside the desktop app, no terminal needed
The approach targets a growing demand: using capable AI without shipping names, addresses or account numbers to third-party servers unprompted. Perplexity’s decision to open-source the classifier makes it independently auditable – how reliably the split holds up in everyday use, however, remains to be seen.
Sources: Perplexity (blog) · 9to5Mac · MarkTechPost
