Perplexity has launched Hybrid Compute, a new feature that splits AI tasks between cloud models and a local model running on a user’s Mac to keep sensitive data on the device.
The feature is built into Perplexity’s Mac app and is aimed at users who want more privacy or lower inference costs by shifting part of a job away from cloud-based models such as Opus 5 or GPT-5.6 Sol.
Perplexity said Hybrid Compute can be used for tasks such as comparing confidential case material with public case law, allowing sensitive client information to remain on the user’s machine while other parts of the task run in the cloud.
Jon Staff, who oversees Perplexity’s Mac products, said the app will automatically check files and information for sensitive content before anything is uploaded. “Any time you try to upload files or send information, we’re going to automatically check for sensitive content and make sure that you want to share that data to the cloud,” he said.
As part of the release, Perplexity trained a privacy classifier that suggests which files or information should stay on the computer. Users can review those decisions before Hybrid Compute assigns work across models.
Users can also choose which models will handle a task. Local options include Gemma E4B and two versions of Qwen’s 35-billion-parameter 3.6 model, including one variant post-trained by Perplexity.
Perplexity said installing a local model does not require terminal access because the Mac app manages the setup process. During processing, the app shows local CPU, GPU and memory usage, alongside a sidebar that tracks token consumption.
Users are not charged for tokens generated by a local model on their own machine. After an output is produced, users can add follow-up instructions or queue tasks from an iPhone.
Staff said fully cloud-based results will usually produce stronger output quality. “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.
He added that some users may place greater weight on privacy and cost than on maximum model performance. “We want the user to have control over where they are on that sliding scale for the particular type of work they want to do,” Staff said.
Hybrid Compute is currently available only on Apple Silicon Macs running macOS 15, and Perplexity recommends at least 32GB of unified memory. The feature is available to Pro and Max subscribers and enterprise customers.
Featured image credit
