Perplexity's Mac app keeps sensitive steps on a local model

Perplexity’s answer to the data-privacy objection is to keep part of the job on your own Mac.

How Hybrid Compute works

Launched on September 1, 2026, Hybrid Compute splits a single task between frontier models in the cloud and compact models running locally. The task starts in the cloud. When a local classifier detects potentially sensitive information, it alerts the user and asks whether to route that step to a local model, or to let the cloud model handle it anyway. The cloud model orchestrates both sides, so the user gets one result rather than two workflows.

The local options at launch are Gemma E4B, Qwen3.6 35B-A3B and a version of Qwen3.6 35B that Perplexity post-trained itself. Setup is a one-click model download. A task can also begin on an iPhone or iPad and hand its sensitive steps to a Mac that is switched on and running the app.

Who gets it

Hybrid Compute is available to Pro, Max and Enterprise subscribers on Apple silicon Macs. The company describes it as the first hybrid inference orchestrator of its kind.

Why it matters

Enterprise buyers have spent two years asking whether documents sent to an AI assistant leave the building. Perplexity’s design does not answer with a policy; it answers with a routing decision the user can see and override. That is a different kind of guarantee than a data-processing agreement, and it depends on the classifier catching what it should. The next question for corporate IT is how that classifier performs on their own documents, which no launch post can answer.