Hybrid agentic inference is coming soon to Perplexity Computer: What is it

Perplexity has introduced a new AI feature called hybrid agentic inference for its Personal Computer platform, which automatically splits tasks between on-device and cloud models. The system is designed to enhance privacy by keeping sensitive data—such as financial records, health information, or private documents—local while offloading computationally intensive tasks to cloud-based AI models. This approach aims to balance efficiency, performance, and user control over data, reducing reliance on centralized cloud infrastructure. The technology addresses what Perplexity calls the 'orchestration problem,' where AI systems must decide which workloads run locally and which require cloud processing. For example, analyzing a bank statement could involve handling sensitive data on the device while relying on cloud models for complex reasoning. The system eliminates the need for manual user intervention by dynamically routing tasks based on their requirements. Perplexity emphasizes that modern PCs, with advancements in processors and AI hardware, are increasingly capable of handling more workloads locally. The company argues that users prefer controlling their data within their devices rather than relying on external servers. This shift aligns with broader industry trends toward decentralized AI processing, reducing dependency on large cloud data centers. The announcement was made in partnership with Intel, with support for multiple hardware platforms, including NVIDIA’s RTX Spark. Perplexity states that the hybrid approach allows local and cloud models to collaborate, optimizing performance while minimizing data exposure. The feature is expected to improve privacy, reduce cloud compute costs, and give users greater autonomy over their AI interactions.