Perplexity adds hybrid cloud-and-local AI processing to its Mac app
Perplexity’s Hybrid Compute mode for Mac divides work between cloud and on-device models, with local controls deciding how sensitive data is handled.
Perplexity has released Hybrid Compute for its Mac app. The feature lets its Perplexity Computer agent divide a task between cloud models and a model running on the user’s Mac. Perplexity says it is available in the newest app release for Pro, Max and Enterprise subscribers.
The mode requires an Apple silicon Mac running macOS 15 or later with at least 24GB of unified memory. Perplexity says cloud models handle web search, planning and frontier reasoning. A local model processes private files, sensitive information and actions performed on the device, while Computer coordinates the workflow.
Before information leaves the Mac, Perplexity says an on-device classifier looks for sensitive details such as names, addresses, account numbers and secrets. Its classification feeds a privacy gate that determines whether protected information stays local, is masked or redacted before cloud processing, causes an action to be refused, or requires the user to approve sending selected context to a cloud model.
Credentials, payment-card numbers and government IDs receive the strictest handling, according to the company. The app can keep that data on the Mac, refuse the action or rewrite the cloud request so processing continues without the protected information.
Perplexity says Enterprise administrators can set organization-wide rules for what stays on a Mac, what may be masked before cloud use and what requires user approval. The company also says administrators can audit when information leaves a device.
At launch, Perplexity lists three local-model options: Gemma 4 E4B, Qwen3.6 35B-A3B and a Perplexity model. Its related PII-TRACE research describes the company-built classifier behind the on-device detection approach.
More news

AWS releases six open-source Hugging Face deployment skills for SageMaker

Google Research releases MilleMiglia logistics benchmark generator

AWS launches AgentCore Runtime V2 with elastic memory and snapshot starts
