Perplexity Launches Hybrid Compute for Secure Local AI Processing

California, USA — Ekhbary News Agency

Perplexity has unveiled Hybrid Compute, a novel tool designed to enhance data privacy and manage costs by distributing AI tasks between powerful cloud models and local large language models (LLMs) running directly on a user's machine. This innovation allows for the secure processing of sensitive information locally, preventing its transmission to external servers.

Balancing Confidentiality and Performance

The new system integrates directly into the Mac app, automatically scanning for sensitive content before suggesting what data should remain on the user's computer. Users retain control, able to review files slated for local processing and select their preferred local LLM, including options like Gemma E4B or Qwen's 35-billion parameter models. Jon Staff, who oversees Perplexity's Mac products, emphasized the system's ability to keep client data confidential, citing a lawyer preparing a brief as a prime example. For what it's worth, this approach directly addresses a critical concern for professionals handling proprietary or personal information.

Cost Efficiency and User Control

Beyond privacy, Hybrid Compute offers a practical solution for reducing inference costs. By offloading less demanding portions of a task to a local model, users can circumvent the typically higher expenses associated with frontier, cloud-based AI. Staff acknowledged that while fully cloud-based systems generally yield superior raw output, the balance between cost, capability, and data privacy is a sliding scale. Perplexity aims to empower users to define their optimal point on this spectrum, ensuring the solution aligns with their specific workflow and budget. The feature currently requires Apple Silicon Macs running macOS 15 with at least 32GB of unified memory and is available to Pro, Max, and enterprise subscribers.