Perplexity Computer will usher in an important upgrade this summer: support for automatically allocating tasks between local and cloud AI models to achieve hybrid reasoning capabilities that better balance privacy protection and computing performance.
Perplexity says Perplexity Computer is the company’s agent system that “does real work for users” by automating the orchestration of different models and tools to handle complex tasks. New features coming soon will allow the system to intelligently switch between a compact model that runs locally on the device and a more powerful, cutting-edge model in the cloud, depending on the type of task and data sensitivity. According to the official introduction, this design aims to stay as local as possible when processing sensitive data, and at the same time, it can seamlessly call the cloud model when strong computing power is needed, thereby striking a balance between security and efficiency.
According to reports, under the new "hybrid agentic inference" mode, work involving sensitive content such as financial records, health information, and personal files will be prioritized by a local small model running on the user's device to determine which data must be strictly retained for local processing. At the same time, tasks that do not require high privacy but require stronger reasoning and large-context capabilities will be automatically assigned to cutting-edge models in the cloud for execution. Perplexity emphasizes that most real-world tasks often contain both sensitive and non-sensitive parts, so the system automatically splits the task into several subtasks and coordinates scheduling between local and cloud, rather than requiring users to choose between "local" or "cloud" mode in advance.
Perplexity further demonstrated the concept of this hybrid architecture in a video released on social platform According to official sources, the hybrid AI orchestration feature is scheduled to be officially launched to Perplexity Computer users in July.