Artificial intelligence company Perplexity has recently announced that it will introduce a major upgrade to its intelligent agent system, Perplexity Computer, this summer. The system will officially incorporate a new "hybrid agent reasoning" mode, which can intelligently switch and automatically allocate between compact models running locally on the device and cutting-edge models in the cloud, based on the type of task and the sensitivity of the data.
As an intelligent assistant that performs actual complex tasks for users, the core of Perplexity Computer's upgrade lies in balancing privacy protection and computing performance. When the system receives a task, it will automatically split it into multiple subtasks and coordinate scheduling between local and cloud environments, thus avoiding forcing users to choose between security and speed.

Sensitive data remains strictly local
In the new mode, tasks involving highly sensitive content such as financial records, health information, and personal documents will be prioritized for processing by local small models running on the user's device. These local models will strictly control access, ensuring that core private data never leaves the individual device, effectively eliminating the risk of privacy leaks from the source.
At the same time, complex tasks with low privacy requirements but high demands for logical reasoning, large context windows, and strong computing power will be automatically sent to the cloud's advanced large models for processing. This design ensures both privacy protection and maximum utilization of the cloud's top-tier computing power and token efficiency.
Collaborative evolution breaking traditional models
Notably, Perplexity emphasizes that most tasks in the real world often contain both sensitive and non-sensitive information. Therefore, Perplexity Computer's automatic splitting and collaborative orchestration capabilities are particularly important, making the entire reasoning process transparent and smooth for users and providing a seamless hybrid AI experience.
According to the official plan, this hybrid AI orchestration and reasoning feature is expected to be released to Perplexity Computer users in July this year. This not only demonstrates the great potential of the edge-cloud integrated architecture but also sets a new benchmark for the integration of future AI agents in daily work and privacy protection.
