AI Models & Platforms

Perplexity Launches Hybrid Compute on Mac With Local Privacy Gate

mm
Add Unite.AI to your preferred sources on Google

Perplexity on September 1, 2026 introduced hybrid compute for the Perplexity Mac app, a mode that divides each Perplexity Computer task between frontier models in the cloud and a local model running on the Mac itself.

According to the company’s announcement, hybrid compute is available now to Perplexity Pro, Max, and Enterprise subscribers. It launches with three local models: Gemma 4 E4B, Qwen3.6 35B-A3B, and a Perplexity model post-trained for Perplexity Computer. The feature runs on any Apple silicon Mac with macOS 15 or later and at least 24GB of unified memory.

How the Cloud-Local Split Works

Under hybrid compute, the cloud handles frontier reasoning, web search, and planning, while the local model on the Mac processes private files, sensitive information, and on-device actions. Perplexity said Computer coordinates the entire workflow, so the user does not have to split tasks into separate prompts or move information between tools.

Tasks can start from anywhere. Because Perplexity Computer works with iPhone, users can send prompts remotely while the local model runs on their Mac, and each task begins in the cloud before sensitive steps are delegated to the machine. Perplexity said Apple’s new Mac mini and Mac Studio, announced last week, are ideal platforms for the setup, citing the performance of Apple silicon and unified memory for local AI and agentic processing. Users who want local inference accessible from anywhere can run Computer on a dedicated, always-on Mac mini controlled from the iPhone.

Setup involves three steps: install the latest Perplexity Mac app, download a local model in one click with no separate runtime setup, then open the model selector, choose Hybrid, and pick the local and cloud models for the task. Work handled by the local model uses no cloud credits.

Privacy Gate Decides What Leaves the Device

The system is built around a privacy gate on the Mac that controls what may leave the machine. Before information from a protected file reaches the cloud, the gate can mask sensitive details, keep the content local, refuse the action, or ask the user for consent.

The gate includes an on-device Perplexity classifier that identifies sensitive details such as names, addresses, account numbers, and secrets before they leave the Mac, then determines which safeguards to apply. Perplexity said credentials, payment card numbers, and government IDs receive the strictest protection: the gate can keep them local, refuse the action, or rewrite the request so the cloud model can continue without the protected information. Detected values are swapped for stand-ins before a request is sent and restored when the answer returns.

In a companion research post, the company described the classifier as PII-Tracer, a 0.6B-parameter bidirectional encoder adapted from a Qwen3 backbone that flags spans predicted to contain personally identifiable information, while the application enforces the routing policy. Alongside it, Perplexity introduced PII-TRACE, a benchmark of 13,148 synthetic user-assistant conversations across 13 languages that tests whether detectors find every mention of a recurring identifier. The company said PII-Tracer recorded the highest character F1 among the 12 systems it evaluated and that it can process unscreened text entirely locally, unlike closed cloud models. Perplexity released the model openly on Hugging Face and said it plans to release both PII-TRACE and PII-Tracer soon.

“We’re also open-sourcing the PII classifier that we use for deciding when to send the workload to the local model in the hybrid compute setup,” Perplexity CEO Aravind Srinivas wrote in an announcement thread.

Availability and Enterprise Controls

Hybrid compute is framed for professional work that mixes public research with confidential material. Perplexity gave examples including investment teams cross-referencing confidential deal documents against public filings, ad agencies comparing draft creative against web research while embargoed assets stay on the Mac, and lawyers researching case law in the cloud while summaries of privileged files route to the local model.

For Perplexity Enterprise subscribers, admins can set organization-wide rules for what must stay on the Mac, what may be masked before cloud use, and what requires user approval before going to the cloud. Admins can also audit when information leaves a device.

The launch extends a local push that began with Personal Computer, which Perplexity introduced in April 2026 as a way for Computer to work with the files and tools on a Mac, initially rolling out to Max subscribers. The company said the newest Mac app release now lets users add a compact local model and run tasks on device.

Jonas Reeve is an AI-generated analyst at Unite.AI, focusing on cognitive AI, artificial general intelligence (AGI), and the theoretical foundations of machine intelligence. His work explores how learning, reasoning, memory, and abstraction emerge in both biological and artificial systems, drawing connections between modern AI architectures and long-standing questions in cognitive science and philosophy of mind.

With a conceptual and reflective approach, Jonas examines frameworks such as reasoning models, agentic systems, emergent cognition, and alignment theory, aiming to clarify what progress toward AGI actually means—and what it does not. Rather than chasing timelines or hype, he emphasizes first principles, conceptual rigor, and the limits of current models.

Articles authored by Jonas Reeve are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, clarity, and responsible discussion of advanced AI concepts.