Perplexity Computer can now run entirely on Apple Silicon chips without sending data to the cloud

Local AI is on everyone’s lips. We have already seen NVIDIA’s insistence on this with RTX Spark, with Microsoft’s Surface Ultra as the main exponent. Well, now it’s Perplexity’s turn. This company, known mainly for its browser and search engine, has made a move.

It does this by updating Perplexity Computer and allowing it to run locally on Macs that have Apple Silicon chips. How exactly does it work? Basically, it moves both the language model and the execution environment and task automation to the computer itself. Thus, the main premise of local AI is fulfilled, which is that personal and corporate files are not uploaded to the cloud.

In reality, this tool arrived at the beginning of the year as a cloud platform capable of working with different cloud models. In September it introduced a hybrid mode that combined both remote and local AI.

Now, hePro, Max and Enterprise plan users They can interact with applications from the Apple ecosystem, such as Notes, Messages, Mail or Numbers, and manage documents without connecting to external servers and without paying for computing. By the way, an ecosystem, both hardware and software, favorite among AI laboratories when testing their new projects.

The technical ins and outs of Perplexity Computer

On a more technical level, we have to tell you that the local version of Computer allows you to select between Qwen 3.6 and PPLX Proan optimized variant of said model after subsequent training. To execute these complex tasks locally, the company recommends having at least 36 GB of unified memoryciting configurations with M5 Pro or M6 chips as reference.

It is important to emphasize that all generated code and system tools operate in an isolated environment or sandboxwhich restricts unauthorized access to other directories on the file system.

In practice, this scheme allows complete offline workflows to be solved. For example, the agent can review an investment portfolio by comparing local documents with guidelines downloaded in an Excel workbook and generate a presentation in Keynote without exposing financial statements. Likewise, it is capable of collecting project requests from the Messages application, checking budget items and preparing email drafts ready for user supervision.

In any case, the macOS application does not give up external infrastructure. The user can toggle between strictly local execution or Hybrid Compute which we were talking about before. If the workflow is complex, the system requests authorization to scale to the cloud and access more advanced models.