Local AI Agent Arrives on Windows with Hardware Requirements

Perplexity has expanded its Portable Computer platform to Windows systems, marking a significant shift in how consumers can access advanced AI capabilities on their own machines. Previously limited to Linux environments, the local AI agent is now available through the standard Perplexity Windows application. This development gives Windows users the ability to run sophisticated multistep AI tasks directly on their hardware without constant cloud dependency.

Portable Computer functions as a local version of Perplexity Computer, which handles complex, multistep assignments rather than simple question-and-answer interactions. The agent can plan tasks, execute subtasks through various connectors and tools, and deliver results beyond basic chat responses. What sets the local version apart is that processing happens on your own system, which means tasks do not consume Perplexity credits. The agent retains the option to send work to cloud models with your explicit permission if needed, but local processing remains the default and primary benefit.

GPU Requirements Define Compatibility

graphics card installation computer hardware
Photo by Thomas Foster

Not all Windows PCs will support this technology. Perplexity and Nvidia have established a clear hardware threshold: your system must have at least 24GB of video RAM and use an Nvidia GeForce RTX or RTX PRO graphics card. This VRAM requirement spans multiple GPU generations, making the specification about memory capacity rather than a specific product generation.

Qualifying consumer GPUs include the RTX 3090 and 3090 Ti (both with 24GB), the RTX 4090 (24GB), and the newer RTX 5090 (32GB). On the professional side, RTX PRO Blackwell cards such as the 4000 (24GB), 4500 (32GB), 5000 (48GB and 72GB options), and 6000 (96GB) all meet the requirements. Users with these cards already installed in their systems are positioned to adopt Portable Computer immediately. Those without qualifying hardware face a choice: purchase a separate AI device or upgrade their current desktop with a used compatible GPU card, which may cost less than standalone solutions.

New Features and Practical Advantages

The Windows release introduces several new capabilities designed for everyday users. Scheduled recurring tasks allow you to set AI agents to handle repeated work automatically on your own timeline. Local MCP (Model Context Protocol) servers enable direct integration with desktop applications including Microsoft Word, Google Drive, Gmail, Slack, and GitHub. These connectors let the AI agent interact with your existing tools and workflows without routing data through external servers.

Setup has been simplified considerably. The application includes a one-click download option for local models, with Qwen 3.8 27B offered as an example starting point. Nvidia continues advancing its GPU technologies for AI applications, and the Portable Computer rollout reflects this momentum toward practical local processing.

Subscription and Cost Considerations

AI agent workflow task management
Photo by kenny cheng

Running Portable Computer requires a Pro or Max Perplexity subscription. This means the software itself is not free, though users with qualifying GPUs avoid the per-task credit costs that cloud-based AI services impose. For power users who generate significant AI workloads, the subscription model combined with local processing can reduce ongoing expenses compared to traditional cloud AI services that charge per request or per credit.

Desktop owners without compatible GPUs have another option. Nvidia and its partners including Lenovo and Acer are releasing RTX Spark Windows PCs expected in October. These prebuilt systems come with the necessary hardware already integrated. The DGX Spark Founders Edition, positioned as a premium all-in-one AI workstation, carries a $4,699 price tag. For existing PC owners, upgrading with a used qualifying graphics card represents a more budget-friendly path to entry.

What This Means for Buyers

This development democratizes local AI access for Windows users who already own compatible hardware. The elimination of cloud dependency for processing reduces latency and keeps your data on your own machine. GPU compatibility continues evolving across different vendors, but this Nvidia and Perplexity partnership specifically targets Nvidia RTX owners.

If you own an RTX 3090, 4090, 5090, or compatible professional RTX card with 24GB or more VRAM, you now have the option to subscribe to Perplexity and run local AI agents for multistep tasks without cloud reliance. For those without qualifying graphics cards, the decision hinges on whether upgrading your system or purchasing a dedicated AI PC aligns with your budget and workflow needs. The expanding ecosystem of local AI tools and agents suggests this category will continue growing, making current investments in compatible hardware increasingly valuable.