NVIDIA and Perplexity Bring Local AI Agents to Windows RTX PCs

The catch? You’ll need a graphics card with at least 24GB of VRAM — a specification that puts the feature out of reach for most everyday PC users.

NVIDIA has announced a collaboration with Perplexity AI that brings the latter’s Portable Computer product to Windows PCs running compatible NVIDIA GeForce RTX and RTX PRO GPUs. The partnership marks the next stage of a rollout that began on Linux, where Portable Computer first launched on RTX GPUs and NVIDIA’s DGX Spark hardware. Windows support is now the latest expansion of that rollout.

What Is Portable Computer, Exactly?

Perplexity’s Portable Computer is the company’s local-first AI agent product. Rather than sending every task to a remote server, it runs the agent harness, workflows, and underlying models directly on the user’s machine. Cloud-based frontier models remain available as a fallback when needed, but the default is to keep processing on-device.

That distinction matters. Running AI workloads locally means sensitive files, prompts, and outputs don’t need to leave the device for routine tasks. NVIDIA states that local execution keeps sensitive information on the PC and avoids sending tasks to the cloud for local workloads — a claim that will resonate with anyone who’s thought twice before pasting a confidential document into a browser-based AI tool.

The Hardware Barrier

The 24GB VRAM requirement is the number that defines who can actually use this. Most consumer RTX cards — including popular mid-range options like the RTX 4070 — ship with 12GB of VRAM. Cards that meet the 24GB threshold include the RTX 4090, RTX 6000 Ada, and RTX PRO 6000 Blackwell, all of which sit at the upper end of the consumer and professional GPU market.

So this isn’t a feature for the typical Windows laptop or budget desktop. It’s aimed squarely at high-end workstations, creative professionals, developers, and researchers who already have the hardware to match. For everyone else, the cloud fallback remains the practical option.

From Linux to Windows: A Staged Rollout

The progression from Linux to Windows is a familiar pattern in the local AI space. Linux tends to attract early adopters, developers, and technical users who are comfortable configuring their own environments. Windows support broadens the potential audience considerably — Windows remains by far the dominant desktop operating system globally.

By launching on Linux first, Perplexity and NVIDIA were able to test the system with a technically engaged user base before committing to the wider Windows rollout. That’s a reasonable approach, though it does mean the Windows version arrives after the initial wave of attention.

The Bigger Picture: On-Device AI Is Gaining Ground

This launch is part of a broader industry shift towards on-device inference. The argument for running AI locally isn’t just about privacy — it’s also about latency, cost, and independence from cloud subscription models. When a model runs on your own GPU, you’re not burning through API credits or waiting on a remote server.

Aravind Srinivas, chief executive of Perplexity AI, said: “We want AI to work for you wherever you are — on your device, with your data, on your terms.”

That framing reflects where the industry is heading. Companies including Apple, Google, and Microsoft have all pushed on-device AI features over the past year, though the specific capabilities and hardware requirements vary considerably between platforms. Perplexity’s approach — pairing local execution with cloud fallback — sits somewhere between a fully offline model and a purely cloud-dependent service.

What the Collaboration Means in Practice

For users who do meet the hardware requirement, the setup is straightforward in principle: install the Perplexity app on a compatible Windows PC, and Portable Computer runs the agent stack locally. Tasks that exceed local capacity, or that benefit from more powerful frontier models, can still route to the cloud. It’s a hybrid model, not an all-or-nothing proposition.

NVIDIA’s involvement is partly about showcasing what RTX hardware can do beyond gaming and content creation. The company has been pushing RTX PCs as capable AI workstations for some time, and partnerships like this one give that positioning concrete substance. An AI agent that runs privately on your own machine is a more compelling demonstration than a benchmark score.

But the 24GB VRAM floor remains a real constraint. Until that requirement drops — either through software optimisation or the arrival of more affordable high-VRAM cards — Portable Computer on Windows will remain a niche proposition.

What This Means for Kent Residents

Kent residents with a high-end NVIDIA RTX GPU — such as the RTX 4090 or a professional RTX PRO card with 24GB or more of VRAM — can download the Perplexity app and use Portable Computer on Windows today. For the majority of PC users in the county, the hardware requirement means this particular feature isn’t yet within reach, though the cloud fallback keeps Perplexity’s broader AI tools accessible to anyone with a standard internet connection. As on-device AI continues to develop, future software updates or more affordable high-VRAM cards could lower the barrier for everyday users across the UK.