Perplexity has brought its local AI agent Portable Computer to Windows systems, partnering with Nvidia to enable offline task automation. Yet a steep hardware requirement blocks the majority of desktop owners: the host machine must carry an Nvidia GeForce RTX or RTX PRO card packing at least 24GB of video memory.
Release Details and Requirements
Portable Computer originally launched for Linux on Aug. 25. It acts as a local copy of Perplexity Computer, the company's cloud agent that can plan, run subtasks through connectors and produce outputs beyond simple chat responses. Running locally does not consume credits, although the agent can forward requests to cloud models with user permission. A Sept. 3 post by Nvidia confirmed Windows support was imminent, and the company listed connectors for Microsoft Word, Google Drive, Gmail, Slack and GitHub.
The high VRAM threshold cuts across Nvidia's consumer lineup. Cards that meet the 24GB minimum include:
The company did not explicitly confirm whether the RTX 5090 Laptop GPU at 24GB qualifies. Nvidia expects RTX Spark Windows PCs from Lenovo and Acer to ship in October, which would offer another route into this capability.
Cost Dynamics and Accessibility
Users who already own a qualifying RTX-equipped desktop now gain access without needing a separate device. Upgrading an existing system with a used qualifying card could cost far less than buying Nvidia's DGX Spark Founders Edition, which recently jumped to $4,699. This shift places local AI within reach of enthusiasts willing to invest in high-end gaming hardware rather than specialized machines.
The requirement effectively gates the technology behind enthusiast-tier silicon. Most consumer laptops lack the VRAM, and even many desktop builders consider 24GB excessive for traditional gaming. Nvidia has positioned these high-capacity cards as engines for generative AI workloads, and the partnership reinforces that strategy.
Why This Matters
This launch advances the vision of fully private, offline AI assistants that do not rely on internet connections or incur per-task costs. However, the hardware bar limits mainstream adoption. For everyday consumers, the value proposition remains unclear until similar tools run on integrated graphics or mid-range cards. The real impact may emerge when Nvidia and Perplexity bring lower-VRAM versions to market, expanding beyond the current niche of power users with $1,000+ graphics cards.
Developers and early adopters stand to benefit most today. They can build workflows that combine local models with cloud fallbacks, gaining flexibility while controlling sensitive data. The Portale Computer for Windows release marks a step toward practical local intelligence, but the VRAM wall ensures that step remains small for now.



