Guide from the official docs

Can you run Perplexity Computer locally? Mac, Windows and Linux requirements

Perplexity Computer can now run tasks on your own Mac, PC or DGX Spark. The plans, memory, models and disk space Perplexity lists, and what still needs the cloud.

By AideDex desk · published

This guide follows the official docs and announcements, linked inline. AideDex did not test these products and does not score them. How we check facts.

Last checked Oct 27, 2026.

Yes, on supported hardware and a paid plan. Since Oct 20, 2026, Perplexity Computer can run “the AI model, harness, and task execution entirely on device” on Apple silicon Macs, and Perplexity recommends 36 GB of memory or more for that (Perplexity). On Windows and Linux, a feature called Portable Computer runs the model and agent on a PC with an NVIDIA RTX GPU with at least 24 GB of video memory, or on an NVIDIA DGX Spark (Perplexity Help Center). Local runs need a Pro, Max or Enterprise subscription, and work the local model does doesn’t use Computer credits (Perplexity Help Center).

Web searches, connected apps, browser use and frontier models still run in the cloud, and only when you allow it. For Computer’s status and plans, see the Perplexity Computer agent page. Comet, Perplexity’s browser with its own Assistant, is a separate product; its setup is in how to get started with Perplexity Comet. Everything below comes from Perplexity’s own pages; AideDex has not run these models.

Three ways Computer can use your machine

Perplexity uses three names that are easy to mix up:

NameWhat runs on your deviceWhere
Personal ComputerThe desktop app gives Computer access to your local files and apps; the agent itself can still run in the cloudMac (macOS 15+) and Windows 10 or 11
Hybrid ComputeA local model handles private files and sensitive steps; cloud models plan, search and reasonApple silicon Mac
Fully local (Mac) / Portable Computer (Windows, Linux)The model and the agent both run on your deviceMac; RTX PCs; DGX Spark

Sources: Personal Computer help, Portable Computer help, Perplexity blog.

Perplexity says the Mac app “offers two ways to run Computer”: Hybrid Compute starts in the cloud, while running locally “starts on the Mac, keeping the task’s planning, reasoning, and file processing on device” (Perplexity).

Hardware and plan requirements

PlatformMemory Perplexity listsLocal modelsPlans
Apple silicon Mac, fully local36 GB or more recommendedQwen 3.6 or PPLX ProPro, Max, Enterprise
Apple silicon Mac, Hybrid Compute24 GB or more of unified memory; 32 GB “gives the best results”Gemma 4 E4B, Qwen3.6 35B-A3B, the Perplexity modelPro, Max, Enterprise
Windows PC, NVIDIA GeForce RTX or RTX PRO24 GB or more of VRAMPPLX 27BPro, Max
Windows PC, AMD Ryzen AI Max Series24 GB or more of GPU-accessible memoryPPLX 27B, Qwen 27BPro, Max
Linux (Ubuntu) PC, NVIDIA RTX24 GB or more of VRAMPPLX 27BPro, Max
NVIDIA DGX Spark (DGX OS)Meets the requirementPPLX 27B, Qwen 27BPro, Max

Sources: Perplexity blog for fully local on Mac; Personal Computer help for Hybrid Compute and the Windows, Linux and AMD rows; Portable Computer help for plans on Windows and Linux, which it gives as “Pro or Max, on a personal or an Enterprise plan”.

PPLX Pro is described as “a post-trained version of Qwen” (Perplexity). The Verge’s headline on the launch put it plainly: “you’ll need a lot of RAM” (The Verge). For scale: Apple’s current 14-inch MacBook Pro starts at 24 GB with the M5 Pro chip, and 36 GB is the configuration with the M5 Max and its 32-core GPU (Apple Support). On Windows, Linux and DGX Spark only one local model runs at a time (Perplexity Help Center).

If your GPU reports less than 24 GB, the on-device models and their download buttons are hidden, and the Local Inference page shows a notice instead (Perplexity Help Center). Without supported hardware you can still use Computer in the cloud.

Disk space. Perplexity lists about 19.6 GB for a first Windows setup with PPLX 27B (including the inference engine and classifier), another 19.9 GB to add Qwen 27B, and 38.6 GB for a first DGX Spark setup. Leave extra room: model updates need the new model’s size plus about 5 GB (Perplexity Help Center).

How to set it up on a Mac

For a fully local task, per Perplexity’s launch post (Perplexity):

  1. Download and open the latest Perplexity Mac app from perplexity.ai/personal-computer.
  2. Follow the in-app instructions to download and set up a local model.
  3. Choose fully local to run the task on the Mac.

For Hybrid Compute instead, open the model selector, select Hybrid, then choose the local model and the cloud model for the task (Perplexity Help Center). When you give it permission, Computer can work with local files and Mac apps including Notes, Messages, Mail and Numbers (Perplexity). Engadget’s example for fully local mode: with access to Notes and Mail, Computer can draw up a priority to-do list without the contents of your emails going to the cloud (Engadget).

How to set it up on Windows or Linux

Portable Computer came to Windows on Sept 14, 2026, after DGX Spark and Linux RTX PCs; NVIDIA lists connectors for Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub, and says support for its DGX Station is “expected to come soon” (NVIDIA). On Windows there’s no separate installer: downloading a model sets up local inference (Perplexity Help Center).

  1. Install the Perplexity app for Windows from the Microsoft Store or Perplexity’s download page, and sign in. On Ubuntu or DGX OS, install the perplexity package from Perplexity’s apt repository.
  2. Open the Menu, select Settings, then Local Inference.
  3. Under On-device models, download PPLX 27B (or Qwen 27B where offered).
  4. In the composer’s model menu, choose the model under Local.

On Windows, file access is a separate switch: enable Local file system access and turn on Desktop tools in Settings → Permissions (Perplexity Help Center). If your account is managed by an organization, an administrator may need to allow local inference first.

What still goes to the cloud

A local model only knows what’s on your machine. Perplexity lists three things a local task can escalate (Perplexity Help Center):

  • Search: Perplexity Search, wide research and deep research.
  • Reasoning: a choice of 15+ frontier models.
  • Connected apps and browser use.

Portable Computer “asks for your permission before sending anything from your device to the cloud”, and an approved step uses Computer credits at the standard rate (Perplexity Help Center). On the Mac, Perplexity says you can authorize Computer to escalate for current information, browser use and connected apps (Perplexity).

In Hybrid mode, a “privacy gate” on the Mac can mask sensitive details, keep a step local, refuse the action, or ask for your consent before information from a protected file reaches the cloud. Perplexity says credentials, payment card numbers and government IDs get the strictest protection (Perplexity Hybrid Compute).

Other limits from the Portable Computer help page: Brain and saved memory “are not available in local mode”, and the local model “is smaller than a frontier model” (Perplexity Help Center).

Does it work offline?

Partly. Perplexity says local tasks “do not depend on a cloud round trip”, but signing in, downloading the model and inference engine, and any step you escalate all need an internet connection (Perplexity Help Center).

Credits

Work the local model handles “does not consume Computer credits”; any step that runs in the cloud uses credits like other Computer work (Perplexity Help Center). Perplexity pitches this for long or recurring jobs, which you can schedule to run on the device (Perplexity Help Center). Plan prices are on the cost page, and other agents’ limits are in Hit a usage limit on your AI agent?.

Local tools: a caution

Portable Computer can use MCP servers running on your own machine, added under Settings → MCP servers → Add server → Local (stdio). Perplexity warns that local servers “run outside the sandbox”, with your permissions and access to your files and the network, “so only add servers you trust” (Perplexity Help Center). Code and tool execution otherwise runs in a sandbox (Perplexity). For why tools and web pages matter, see Prompt injection: can a web page or email hijack your AI agent?.

Other agents you can run on your own hardware

If you want an agent whose model runs on your machine without a Perplexity plan, the open-source OpenClaw and Hermes Agent can use local models; their memory needs and costs are in OpenClaw and Hermes Agent hosting costs. Agents that keep working with your laptop shut are in Which AI agents keep working when your computer is off?.

FAQ

How much RAM do I need to run Perplexity Computer locally on a Mac?

Perplexity recommends 36 GB or more for running Computer fully local on a Mac (Perplexity). Hybrid Compute needs at least 24 GB of unified memory (Perplexity Help Center).

Is local mode free?

No. It comes with a paid plan, and what it saves is that plan’s credits: MacStories sums it up as letting subscribers “conserve credits, making their subscriptions go further” (MacStories). The other cost is hardware with enough memory.

Sources

9 pages
  1. Perplexity: Perplexity Computer now runs fully local on Mac (Oct 20, 2026) perplexity.ai
  2. Perplexity: Introducing Hybrid Compute on Mac (Sep 1, 2026) perplexity.ai
  3. Perplexity Help Center: What is Personal Computer? perplexity.ai
  4. Perplexity Help Center: What is Portable Computer? perplexity.ai
  5. The Verge: Perplexity Computer can now run fully offline on Mac (Oct 20, 2026) theverge.com
  6. NVIDIA Blog: Perplexity Portable Computer is now available on Windows (Sep 14, 2026) blogs.nvidia.com
  7. Apple Support: MacBook Pro (14-inch, M5 Pro or M5 Max) tech specs support.apple.com
  8. MacStories: Perplexity Computer's local-first approach to agents (Oct 20, 2026) macstories.net
  9. Engadget: Perplexity makes it easier to run AI workflows entirely on your Mac (Oct 20, 2026) engadget.com

Spotted something wrong or out of date? Report an error · Corrections log