Perplexity’s AI agent can now run entirely on your Mac, and it wants 36GB of memory

Perplexity launched a fully local mode for its Computer AI agent on Mac on Tuesday, running the model, the agent harness and the task entirely on the machine and recommending 36GB of memory or more — half again the 24GB minimum Perplexity set for its cloud-and-Mac Hybrid Compute mode in September. The local mode runs on Qwen 3.6 or PPLX Pro, a post-trained version of Qwen, and is available now to Perplexity Pro ($20 a month), Max ($200 a month) and Enterprise subscribers on Apple silicon Macs, with code and tool execution in isolated sandboxes; users can still authorize escalation to the cloud and to more than 15 frontier models for harder reasoning.

Perplexity’s launch image for running Computer fully local on Mac. Image: Perplexity Perplexity’s AI agent can now do a whole job without your data leaving your Mac. The company launched Computer fully local on Mac https://www.perplexity.ai/hub/blog/perplexity-computer-now-runs-fully-local-on-mac on Tuesday, running the AI model, the agent harness that plans and uses tools, and the task itself entirely on the machine. Given permission, it works across local files and Mac apps such as Notes, Messages, Mail and Numbers without sending personal data to the cloud. The catch is the hardware. Perplexity recommends 36GB of memory or more for running Computer locally, half as much again as the 24GB minimum it set for its cloud-and-Mac Hybrid Compute mode last month, so plenty of everyday Macs won’t make the cut. What runs on your Mac Computer is Perplexity’s agent that plans multi-step tasks and carries them out with tools, rather than just answering questions. In the fully local mode it runs on one of two models: Qwen 3.6, from Alibaba’s open Qwen family, or PPLX Pro, which Perplexity describes as a post-trained version of Qwen. Users download the model inside the latest version of the Perplexity Mac app, then choose “fully local” when they start a task. Perplexity gives a sense of what that looks like in practice. Local Computer can read a project’s to-do list in Notes, find a teammate’s request for a budget update in Messages, pull planned and actual spending from an Excel file, then write a status update and send a meeting invitation to discuss it. Its other examples include checking private investment holdings against downloaded market data and turning confidential client notes and a rate sheet into a proposal. Code and tool execution run in isolated sandboxes with controlled access to files and connected apps, the company says, the same protections its cloud version uses. That matters on a Mac, where Apple has been clamping down on how much of the system AI agents can see https://madrobot.blog/2026/10/02/apple-macos-full-disk-access-ai-agents-muse-new-controls/ . Local by default, the cloud when you allow it Fully local doesn’t have to mean cut off. Users can authorise Computer to escalate to the cloud for current information, browser use and connected apps, and the local orchestrator can call on any of more than 15 frontier models for harder reasoning. That makes it the mirror image of Hybrid Compute https://www.perplexity.ai/hub/blog/introducing-hybrid-compute-on-mac , which Perplexity launched on September 1. Hybrid tasks start in the cloud and hand private files to a local model; fully local tasks start on the Mac and reach out only when asked. Perplexity pitches the new mode as giving users more control over sensitive data and over their spend on Computer credits, the allowance that cloud tasks draw on. The company announced the launch on X: Perplexity Computer can now run fully local on your Mac. Work with your native Mac apps and local files, and run tasks without sending sensitive data to the cloud. The model and the agent both run on your Mac. Available now in the Perplexity Mac app. October 20, 2026 https://x.com/perplexity ai/status/2112571597798461920 Who can use it Local Computer is available now to Perplexity Pro, Max and Enterprise subscribers in the Perplexity Mac app, on supported Macs with Apple silicon. Perplexity holds up Apple’s new Mac mini as an example machine. Pro costs $20 a month and Max $200 a month in the US. The announcement names no regional limits, and Perplexity sells its paid plans and Mac app internationally, including in the US, UK, Canada, Germany and Australia. Intel Macs, Windows PCs and phones aren’t part of this launch. It is the third step in Perplexity’s push onto the desktop, after Personal Computer for Mac brought the agent to local files and native apps in May and Hybrid Compute added local models in September. It also lands in a busy month for agents that live on your own machine: Microsoft is building a sandbox for AI agents into Windows https://madrobot.blog/2026/10/07/microsoft-windows-copilot-local-files-actions-mxc-agent-sandbox-surface-laptop-ultra/ , and Nvidia’s RTX Spark laptops https://madrobot.blog/2026/10/16/nvidia-rtx-spark-laptops-on-sale-local-ai-agents-prices-models/ ship with installers for agents including Perplexity’s. If you’re shopping for a Mac to run models on, our Mac mini local AI guide https://madrobot.blog/2026/09/23/how-to-run-local-ai-models-mac-mini/ covers how much memory to buy. Why it matters A mainstream AI agent that plans, reasons and acts on your files without touching the cloud is a real shift for anyone wary of sending email, messages or financial data to a server. But the 36GB recommendation means the most private version of Perplexity’s agent is, for now, a feature for people with high-end Macs. Sources: Perplexity: Perplexity Computer now runs fully local on Mac https://www.perplexity.ai/hub/blog/perplexity-computer-now-runs-fully-local-on-mac ; Perplexity: Introducing Hybrid Compute on Mac https://www.perplexity.ai/hub/blog/introducing-hybrid-compute-on-mac ; Perplexity on X https://x.com/perplexity ai/status/2112571597798461920 .