Perplexity launched a hybrid computing function for its Mac application on September 1st that allows an artificial intelligence agent to start a task in the cloud and dynamically transfer the confidential portions of that same task to a model running directly on the user's computer, without restarting the process or losing accumulated context, as reported by VentureBeat. The function, available to subscribers of Pro, Max, and Enterprise plans on Mac computers with Apple Silicon chip, macOS 15 or higher, and a minimum of 24 gigabytes of unified memory, with 32 gigabytes recommended, relies on what the company calls "Privacy Gate": a classifier trained by Perplexity that runs directly on the device and detects personally identifiable information, such as names, addresses, account numbers, and sensitive data, before any information is transmitted to the cloud.
Users can also start a task from an iPhone and transfer local work to their Mac without having to start over, extending the hybrid computing logic beyond a single device. For enterprise customers, Perplexity added configurable company-wide rules on what information must remain local, in addition to an auditable log of what data left each device. The move responds to a tension many companies already face when adopting conversational AI assistants: the most capable models often run in the cloud, but many organizations' privacy and regulatory compliance policies restrict what kind of information can leave their own devices or internal networks.
For companies in regulated sectors, such as financial services or healthcare, which until now had limited their adoption of cloud-based AI assistants due to compliance reasons, the availability of a mechanism that keeps sensitive data on the user's device could reduce one of the most cited barriers to expanding the use of these types of tools in their daily operations.
From next+'s perspective, Perplexity's hybrid computing launch confirms that the next stage of competition among AI assistants is no longer limited to how capable the cloud model is, but to how well it manages to combine that capability with the local processing necessary to comply with the privacy requirements that the most regulated companies demand before adopting any such tool.