Your Mac is now part of Perplexity’s AI infrastructure

Perplexity, the AI startup, has introduced a new feature called Hybrid Compute that allows its AI agents to leverage the computing power of Mac computers running Apple silicon alongside its powerful cloud models. This move comes on the first day of John Ternus' tenure as Apple's CEO, following his leadership in transitioning Apple from Intel processors to its own silicon.

The feature begins with a cloud model handling the initial task. When the task involves sensitive information, the sensitive part of the job is moved to a smaller model running on the user's Mac. This "Privacy Gate" checks for sensitive data and only moves sensitive parts of the task to the local model. Users have control over which model handles the local work and can review what is kept local before starting.

Users can choose between three models for local work, with the option for more in the future. Once the task runs, the app shows local resource usage and token consumption, with no charges for local tokens. This reduces AI inference costs and enhances privacy by keeping more work on the device.

However, moving context from cloud to local models remains a challenge. The cloud model needs to retain enough context to continue the job without accessing the private information that stayed on the Mac. Perplexity's system moves steps between cloud and local models without restarting the task or losing context, but details on context transfer and filtering are not provided.

For enterprise customers, Perplexity offers company-wide rules for local data handling and tracks what leaves each device, addressing similar governance concerns that arise across AI systems. While the Mac hardware requirements for running the AI locally are high, this feature may pave the way for more AI work to move from cloud to device in the future as Macs become more capable.