Perplexity is bringing its local AI agent to Windows PCs. The company has launched Portable Computer for Windows, allowing compatible NVIDIA-powered computers to run complex, multi-step AI workflows locally while keeping files and most of the agent’s activity on the device.

Portable Computer is essentially the local version of Perplexity Computer, the company’s agent that can plan and execute multi-step tasks rather than simply responding to individual prompts. Instead of relying entirely on Perplexity’s cloud infrastructure, Portable Computer moves the model, agent harness, orchestrator, scheduler, and sandboxed execution environment onto the user’s own machine.

That distinction matters because the agent can work with information that users may not want to upload to the cloud. Portable Computer can access permitted documents, spreadsheets, presentations, PDFs, code, and images stored on the Windows PC and use local tools to process them.

Perplexity says work completed by local models does not consume Computer credits, making the setup particularly interesting for repetitive or high-volume tasks. A user could, for example, schedule a workflow to process a collection of files every morning without manually starting the job each time.

An AI agent that actually runs on your PC

Portable Computer goes beyond simply offering a locally running chatbot.

Its local orchestrator can coordinate tasks across files and applications, while the scheduler can keep recurring workflows running on the Windows machine. Perplexity gives the example of a logistics company automatically reconciling freight invoices against local carrier rate sheets, identifying duplicate charges or pricing discrepancies, and producing an exception queue for further action.

The important part is that the underlying invoice and rate information can remain on the company’s own computer throughout that workflow.

Portable Computer can also connect to services including Gmail, Outlook, Slack, and GitHub through the local orchestrator. Other desktop applications can be connected through local Model Context Protocol (MCP) servers, allowing the agent to use locally available tools and data.

That makes Portable Computer considerably more interesting than a conventional local LLM interface. The goal isn’t simply to ask a model questions without an internet connection. It is to give an autonomous agent access to the files, applications, tools, and scheduled jobs that make up an actual computer workflow.

Local by default, cloud when necessary

Perplexity isn’t positioning Portable Computer as an entirely disconnected AI system.

A local model obviously cannot know what’s happening on the internet unless it has access to an external service. Portable Computer can therefore bring in Perplexity Search, wide or deep research, or one of more than 15 frontier cloud models when a task requires current information or more advanced reasoning.

The key difference is that cloud escalation is controlled rather than automatic. If a task needs to send information from the Windows machine to a cloud service, Portable Computer asks the user for permission before proceeding.

For example, someone could keep a confidential document on their PC while asking Perplexity to retrieve current information from the web. The local portion of the task can remain on the computer, while the user decides whether information should leave the device when cloud assistance becomes necessary.

That hybrid approach is becoming one of the more interesting aspects of Perplexity’s agent strategy. Local hardware handles the work it can, while cloud models remain available for tasks that genuinely benefit from them.

Windows brings local agents to a much larger PC ecosystem

Portable Computer arrives shortly after Perplexity brought its broader Personal Computer experience to Windows.

Personal Computer for Windows allows Perplexity Computer to work with local files, Microsoft applications, connected services, and the web. It is primarily a cloud-agent experience with access to the Windows desktop environment. Portable Computer takes the next step by moving the core agent stack and local model onto supported hardware.

The distinction is important. Personal Computer is about giving Perplexity’s cloud-based agent access to the Windows environment. Portable Computer is about putting the agent itself on the Windows machine.

Perplexity’s product page describes Portable Computer as a local-first agent stack that can run tasks locally, schedule recurring work, connect to applications, and then ask before routing a task to cloud models.

You’ll need a serious NVIDIA GPU

There is a fairly significant hardware requirement attached to the Windows release.

Portable Computer requires a Windows PC with a supported NVIDIA GeForce RTX or NVIDIA RTX PRO GPU with at least 24GB of VRAM for local inference. The current Perplexity product documentation lists PPLX 27B as the available local model on Windows RTX PCs.

That means this isn’t an AI feature aimed at the average Windows laptop. A typical integrated-GPU notebook or even a mainstream gaming laptop won’t necessarily qualify.

The 24GB VRAM requirement puts Portable Computer firmly into enthusiast, professional workstation, and high-end AI PC territory. It also explains why NVIDIA hardware is central to the rollout: running a relatively large model alongside an agentic workload requires substantially more local compute than running a lightweight chatbot.

Perplexity says users can set up Portable Computer through its Windows app and download the local model from the model selector. Only one local model runs at a time on Windows.

Recurring AI work is where this gets interesting

The scheduler may ultimately be one of Portable Computer’s most useful features.

Local AI is often demonstrated with simple prompts such as summarizing a document or asking questions about a PDF. Portable Computer is aimed at something more ambitious: workflows that can keep running without someone sitting in front of the computer.

A team could schedule a recurring analysis of files in a permitted folder. An engineer could have the agent review pull requests against a local codebase and prepare a summary. A designer could connect a locally installed application through an MCP server and have the agent work with that application’s tools.

Perplexity says the local orchestrator can also work with GitHub, Slack, Gmail, and Outlook, while cloud research can be incorporated when required.

This is where Portable Computer starts to resemble an AI employee running on a workstation rather than another chatbot living in a browser tab.

There is also a practical economic advantage. Because locally processed work does not incur per-token or Computer-credit charges, workloads involving large collections of local files or repeated processing can be much more practical to run continuously on hardware the user already owns.

Perplexity is betting on the PC as an AI workstation

Portable Computer also shows where Perplexity thinks personal computers are heading.

The company’s earlier Computer product relied heavily on cloud infrastructure and a collection of connected services. Now, Perplexity is building a model where the PC itself becomes part of the AI system: local files, local applications, local models, local scheduling, and local execution can all participate in an automated workflow.

The company first introduced Portable Computer on NVIDIA’s DGX Spark, using local models and an on-device orchestration stack. It then expanded the technology toward NVIDIA RTX hardware, with Windows now joining Linux as a supported PC platform.

The Windows launch could therefore prove more significant than simply adding another supported operating system. Windows has an enormous installed base of PCs, and high-end RTX hardware is already common in gaming systems, workstations, and creator machines.

For now, though, the hardware barrier is substantial. Portable Computer requires at least 24GB of VRAM, and it is available to Perplexity Pro and Max subscribers on individual and enterprise plans.

Still, the direction is clear. Perplexity isn’t asking users to choose between local AI and cloud AI. It is building an agent that can start locally, stay local when possible, and reach into the cloud only when the user allows it and the task actually benefits from additional intelligence or fresh information.

That could make the Windows PC itself a much more capable AI workstation — provided you’ve got the GPU to feed it.