Perplexity's Hybrid Compute splits sensitive tasks between cloud and local AI

Perplexity Back in February, debuted Perplexity Computer. Like Claude Cowork, it's a suite of AI agents that can autonomously complete tasks using the web, as well as files and apps on your PC. Since then, the platform has evolved to encompass a few different products, including Personal Computer for the Mac, and today Perplexity is announcing yet offshoot called Hybrid Compute. The new tools allows you to split a task between a frontier, cloud-based model like Opus 5 or GPT-5.6 Sol and a local LLM running on your computer — the idea being that the local model can handle any sensitive information so that it remains safe and secure on your machine. Perplexity suggests a few different use cases where Hybrid Compute would be a good fit. For instance, a lawyer might want to prepare a brief that compares the case they're working on against existing case law. In that scenario, the tool would allow them to keep their client's data confidential. Perplexity is also pitching Hybrid Compute as a way for thrifty users to save on inference costs by offloading some of the work from the (typically pricier) frontier models. "This is integrated directly into the Mac app. Any time you try to upload files or send information, we're going to automatically check for sensitive content and make sure that you want to share that data to the cloud," said Perplexity's Jon Staff, who oversees all of the company's Mac products. As part of this release, the company trained a new privacy classifier that will automatically suggest files and information the user should keep on their computer. Before Hybrid Compute starts delegating your task between different models, you'll be able to look over what files it wants to gate away to double check it didn't miss something important. At this stage, you can also decide what models you want to tackle the work. On the local side, your options are Gemma E4B and two flavors of Qwen's 35-billion parameter 3.6 model. One of those variants was post-trained by Perplexity. The company says it will offer more local models in the future. Whatever option you choose, installing a local model does not require you to open your Mac's terminal, and Perplexity's app handles the installation process. Once the system is underway, you'll see a visualization displaying your local CPU, GPU and memory usage. Next to that, a sidebar shows how many tokens the task has consumed. You won't be charged for any tokens a local model generates on your personal machine. Once the system generates an output, you can write follow-up instructions as usual. You can also use an iPhone to queue up tasks. I asked Staff if Perplexity benchmarked the outputs Hybrid Compute produced against that of a fully cloud system. "The short answer is that a fully frontier output is going to almost always be better in terms of raw artifact creation. It's more expensive, it's more capable," he said. However, Staff added that some users don't need access to the best, most powerful AI models to do their work, and in those cases, data privacy and cost might be more important factors. "I think it's got to be a sliding scale, and we want the user to have control over where they are on that sliding scale for the particular type of work they want to do. Obviously we will try to suggest the best thing given the context that we have, but ultimately the user needs to be able to decide what solution fits for them," Staff said. For the time being, Hybrid Compute is only available on Apple Silicon Macs running macOS 15, and Perplexity recommends a machine with at least 32GB of unified memory. The feature is available to Pro and Max subscribers, as well as the company's enterprise customers.