Perplexity Launches Hybrid AI to Keep Sensitive Data Local
Perplexity has introduced a hybrid compute feature for its AI platform, allowing tasks to be split between the cloud and local Apple Silicon Macs. The aim is to process confidential data securely on the user's device.

Perplexity has launched a new hybrid compute feature for its AI agent platform, Computer. This system allows a single AI agent to split its workload between advanced models running in the cloud and smaller, open-weight models operating locally on Apple Silicon Macs. The key benefit is routing sensitive data to the local machine, ensuring it never leaves the device.
The company states this is the first time an AI agent can initiate a task in the cloud and dynamically hand off confidential portions to a model running on the user's own hardware, without restarting the job or losing context. This feature is available today through Perplexity's desktop app for enterprise customers who opt-in, as well as for Pro and Max subscribers, on any Apple Silicon Mac running macOS 15 or later.
Perplexity emphasizes the compelling nature of this hybrid approach for tasks where confidentiality is paramount and accuracy is critical. By combining the intelligence of frontier models in the cloud with the security and privacy of local processing, users can achieve better outcomes. The architecture functions like a dispatcher: a cloud-based model breaks down a task into subtasks and routes them appropriately. Web research and heavy reasoning are handled in the cloud, while tasks involving private files or device actions are delegated to a sub-agent running locally on the Mac.
A crucial component is Perplexity's "Privacy Gate," a company-trained classifier that runs on the device. It scans for personally identifiable information before anything is transmitted to the cloud. If sensitive content is flagged, the user can choose whether that portion of the task runs locally or is shared. Locally generated computations are free to the user; only cloud orchestration and delegation consume credits.
Perplexity demonstrated the functionality with professional use cases. One scenario involved a lawyer updating a confidential brief against local files while a cloud agent retrieved public case law. Another showed a private equity associate refining a financial model and generating an investment committee deck. For enterprise clients, administrative tools will be provided for further configuration.