Perplexity Portable Computer for Windows: GPU Requirements, Plans and Local AI

What is Perplexity Portable Computer?

Portable Computer is the local version of Perplexity’s Computer agent. Instead of relying entirely on remote cloud infrastructure, Portable runs a local AI model and major parts of the agent system on hardware you own.

Perplexity says the Windows release brings Computer’s local agent harness and orchestrator to compatible PCs for complex tasks and long-running workflows. The broader Windows experience can combine local files, connected tools, scheduled work, web research and cloud reasoning when needed.

Portable first appeared on NVIDIA DGX Spark hardware. The September 2026 release expands that local-agent approach to supported Windows PCs using eligible NVIDIA GeForce RTX and NVIDIA RTX PRO GPUs.

The important idea is local-first rather than local-only. Sensitive work can remain on the PC, but the same task can selectively call Perplexity Search or frontier cloud models when the user permits it.

Official source: Perplexity — Portable Computer for Windows is here.

Perplexity Portable Computer Windows requirements

Perplexity’s launch announcement gives one hardware requirement very clearly:

That sentence has two separate conditions:

• The GPU must be in the supported NVIDIA GeForce RTX or RTX PRO family.
• The GPU must have 24GB or more of VRAM.

Having an RTX-branded GPU by itself is therefore not enough. A newer GPU can still be incompatible if it has less than 24GB of video memory.

Perplexity’s current Windows launch note does not announce AMD Radeon or Intel Arc support for this on-device inference path.

Requirement sources: Perplexity and NVIDIA Local AI.

Which GPUs qualify? 4090 and 5090 vs 5080

The easiest way to understand the rule is to compare official NVIDIA memory specifications with Perplexity’s 24GB floor.

This is why a search such as “Portable Computer not showing on RTX 5080” can have a simple answer: the 5080 is a powerful RTX GPU, but NVIDIA lists the standard card with 16GB of VRAM, which is below Perplexity’s published local-inference minimum.

Meeting the VRAM threshold does not guarantee that every individual configuration will work perfectly; drivers, the Windows app and other compatibility factors still matter. But failing the 24GB requirement means the PC does not meet the requirement Perplexity currently publishes.

GPU memory specifications: RTX 4090, RTX 5090, and RTX 5080 from NVIDIA.

Why does Portable Computer need so much VRAM?

Portable Computer is not simply using the GPU for a small UI acceleration feature. It runs a capable local model as part of an agent system that can reason across files, invoke tools and stay active through longer workflows.

NVIDIA says the Windows experience simplifies setup around a local model such as Qwen 3.8 27B, post-trained to work with Perplexity Computer and optimized for NVIDIA RTX GPUs.

Large local models need enough graphics memory to hold model weights and runtime state efficiently. Perplexity has therefore chosen 24GB as the current minimum for its supported Windows on-device inference experience.

The practical rule is simple: do not compare GPUs only by generation number or gaming performance. For this feature, VRAM capacity is an explicit eligibility requirement.

Local model example: NVIDIA — Perplexity Portable Computer on Windows.

Which Perplexity plans include Portable Computer for Windows?

Perplexity says Portable Computer for Windows is available to Pro and Max subscribers on both individual and enterprise plans through the existing Perplexity Windows app.

So there are two gates: an eligible subscription and eligible NVIDIA hardware. Paying for Pro or Max does not make a 16GB RTX GPU meet the local-inference requirement.

How to install Portable Computer on Windows

Perplexity does not describe Portable Computer as a separate Windows application. It is available through the existing Perplexity app for Windows.

1. Install or update the Perplexity app for Windows.
2. Sign in with an eligible Pro or Max subscription.
3. Confirm that your NVIDIA GeForce RTX or RTX PRO GPU has at least 24GB VRAM.
4. Open the model dropdown.
5. Select the local model you want to use.
6. Download it with the one-click local-model flow.
7. Choose or grant access only to the local folders and connected tools needed for your workflow.

Perplexity specifically says users can download the desired local model in one click after selecting it from the model dropdown.

What actually runs locally?

Perplexity’s Windows announcement says Portable Computer runs the model, agent harness, orchestrator and scheduler on the Windows device.

That matters because “local AI” can mean very different things. In this case, it is broader than simply opening a file stored on the PC. The local system can plan work, use the selected local model, operate across permitted data and run scheduled workflows without automatically moving that work to a remote model.

Perplexity gives an example of a finance workflow that reconciles invoices against local rate sheets every morning, flags discrepancies and builds an exception queue while the invoice and rate data remain on the company device.

NVIDIA similarly describes Portable Computer as handling multistep work directly on the PC while keeping sensitive information on-device.

Does Portable Computer use Perplexity Computer credits?

Work completed on-device does not consume Perplexity Computer credits. Both Perplexity and NVIDIA make this point explicitly.

This is particularly useful for long-running local workflows. A task that repeatedly processes private files on your own GPU does not need to spend Computer credits simply because it is agentic or scheduled.

Do not overgeneralize that into “Portable can never use credits or cloud resources.” Portable can bring in Perplexity Search or frontier cloud models when a task requires current information or stronger reasoning. Those cloud features remain separate from the statement that locally completed work does not consume Computer credits.

Credit behavior: NVIDIA and Perplexity.

Local does not mean permanently offline

Portable Computer can keep a workflow local when local models and tools are enough, but it can also call cloud capabilities when the task needs them.

Perplexity says users can bring in:

• Perplexity Search for current, cited web information
• 15+ frontier models for advanced reasoning

The privacy boundary is important: Perplexity says that when a task needs to send something from the device to the cloud, Portable asks the user for permission before moving forward.

A useful mental model is:

That is different from a cloud-only assistant that uploads the task context by default.

What local files can Portable work with?

Perplexity says Portable can work with information stored in permitted folders, including:

• documents
• spreadsheets
• presentations
• PDF files
• code
• images

The permitted-folder wording is important. Portable is not described as automatically crawling every file on the Windows machine. The user grants access to the folders needed for the workflow.

Once the information is available locally, the agent can use tools to act on it rather than merely summarize it.

Portable Computer can use local MCP servers

One of the most interesting parts of the Windows release is support for local Model Context Protocol servers running on the Windows device.

Perplexity says this lets the local model use another desktop application’s tools and data as part of an automated workflow while keeping the work on-device.

For example, a locally installed creative or productivity application could expose an MCP server. Portable could then use that application’s tools to create or revise a file while the local model, source files and application interactions remain local.

This produces a useful distinction:

Local MCP support: Perplexity’s Windows announcement.

Which connected apps can it work with?

Perplexity specifically names Gmail, Outlook, Slack and GitHub as services its local orchestrator can work with. NVIDIA’s launch article also highlights connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub.

These connected services are different from permitted local folders. A workflow can combine local context with external systems when the user has connected them.

For example, Perplexity describes an engineering workflow where Portable reviews open pull requests against a local codebase, groups them by status and, after user approval, posts the summary into Slack and tags the responsible person.

Because external connectors can move information outside the local machine, review which service is involved and what the workflow intends to send before approval.

Connector examples: NVIDIA Local AI and Perplexity.

Portable Computer vs normal Perplexity Computer

Perplexity Computer is the broader agent system that can orchestrate models, tools, connected apps and web research. Earlier Windows functionality brought cloud-based Computer agents to the desktop, where they could work with local files, connected apps and the web.

Portable Computer changes where the agent stack can run. It brings a local model plus the agent harness, orchestrator and scheduler onto the Windows PC itself.

Why Portable Computer may not be showing on your Windows PC

If you installed or updated Perplexity for Windows but cannot find the local Portable Computer option, work through these checks in order:

1. Check your subscription. The Windows launch is for Pro and Max subscribers.
2. Check the GPU vendor/family. Perplexity currently names NVIDIA GeForce RTX and RTX PRO.
3. Check VRAM, not just the RTX model number. You need at least 24GB.
4. Update the Perplexity Windows app. Portable Computer ships through the existing app.
5. Update your NVIDIA driver if the app does not detect otherwise-compatible hardware correctly.
6. Look in the model dropdown for the local-model download option.
7. If your hardware and plan meet the published rules but the feature is still missing, use Perplexity support rather than assuming a lower-VRAM card should qualify.

Requirements Perplexity has not clearly published in this launch note

It is useful to separate the requirements Perplexity actually announced from specifications that people may guess.

In the official Windows launch material reviewed for this article, Perplexity clearly states the plan requirement, Windows app, NVIDIA GPU families and 24GB VRAM minimum. It does not provide a complete public minimum-spec table covering every other component.

BeeVaults therefore does not assign invented minimums for:

• system RAM
• CPU model or core count
• free disk space for every supported local model
• a specific minimum Windows 11 build
• a specific NVIDIA driver version
• AMD Radeon support
• Intel Arc support

Local model downloads can be large, so practical disk-space needs will depend on the model Perplexity makes available. Check the app’s download size and live requirements before installing rather than relying on an unofficial fixed number.

What should you do next?

1. Open your GPU specifications and verify VRAM first. If it is under 24GB, the PC does not meet Perplexity’s current on-device requirement.
2. Confirm that the GPU is GeForce RTX or RTX PRO.
3. Make sure your account is on Pro or Max.
4. Install or update the existing Perplexity Windows app.
5. Select and download a local model from the model dropdown.
6. Start with a permitted test folder instead of your entire drive.
7. Run a local-only task and confirm that the output behaves as expected.
8. If a task requests cloud escalation, review what information would leave the device before approving it.
9. Use local MCP only with desktop applications and servers you trust.