Perplexity has started rolling out its Portable Computer feature to Windows PCs equipped with NVIDIA RTX graphics cards, giving everyday users a way to run agentic AI tasks locally instead of routing everything through the cloud. The update, detailed in a September 14, 2026 announcement, means the Perplexity Portable Computer tool now works on NVIDIA GeForce RTX and RTX PRO Workstation GPUs with at least 24GB of VRAM, extending a capability that previously lived only on NVIDIA DGX Spark systems and Linux-based RTX machines.
Key takeaways
- Perplexity Portable Computer is now live on Windows PCs running NVIDIA GeForce RTX or RTX PRO GPUs with 24GB or more of VRAM.
- The tool is a local version of the Perplexity Computer AI agent, capable of handling multistep tasks without sending sensitive data off-device.
- It runs on a local model called Qwen 3.8 27B, which has been optimized specifically for NVIDIA RTX hardware.
- Users can escalate complex requests to cloud-based models when extra reasoning power is needed, but only with explicit permission.
- Connectors link the agent to Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub for real workflow tasks.
Perplexity Portable Computer Launches for Windows PCs with NVIDIA RTX GPUs
The headline change is straightforward: Windows users with the right hardware can now run Perplexity Portable Computer directly on their machines. Perplexity confirmed the feature is available for compatible NVIDIA GeForce RTX PCs and NVIDIA RTX PRO Workstations, provided the GPU carries a minimum of 24GB of VRAM. That threshold matters because it separates casual gaming setups from the higher-end configurations capable of running local large language models at usable speeds.
This isn’t Perplexity’s first attempt at bringing agentic AI outside the browser. The company had already supported NVIDIA DGX Spark systems and RTX PCs running Linux before extending the same tooling to a much larger Windows user base. Windows remains the dominant desktop operating system, so opening the door to NVIDIA RTX AI hardware already installed on millions of consumer and workstation PCs meaningfully widens who can actually use the tool without buying specialized equipment.
Why the VRAM requirement matters
Running a language model locally instead of in a data center demands enough memory to hold the model and its active computations at once. The 24GB floor Perplexity has set effectively points users toward NVIDIA’s higher-tier RTX and RTX PRO cards rather than entry-level graphics hardware, which explains why this rollout targets workstation-class GPUs first.
Technical Features and AI Model Optimization
Portable Computer works as a local counterpart to Perplexity Computer, the company’s existing AI agent designed to plan and execute multistep tasks. Rather than relying entirely on remote servers, this version processes work directly on the user’s machine, drawing on the graphics card’s processing power to analyze files, pull together scattered information, and handle repetitive tasks automatically.
The engine behind that local processing is a model called Qwen 3.8 27B, which Perplexity says has been post-trained and tuned specifically to run efficiently on NVIDIA RTX GPUs. That optimization is what allows the app to skip the technical setup typically required for local AI — users don’t need to hunt for compatible models or wrestle with configuring a software stack, since the app handles that groundwork itself.
Perhaps the most practical design choice is how the system blends local and cloud processing. When a task exceeds what the on-device model can reasonably handle, Portable Computer can hand it off to more powerful cloud-based models for deeper research or reasoning — but only after asking the user for permission first. That built-in checkpoint gives people direct control over when their data actually leaves the device.
Privacy and Integration Benefits of Local AI Processing
Keeping data on the device is the core selling point of running local AI models instead of cloud-only agents. Because the analysis happens on the user’s own hardware, sensitive files never have to touch a remote chatbot or server unless the user explicitly approves an escalation to the cloud. That distinction matters for anyone handling financial records, legal documents, or proprietary business data.
There’s also a cost angle. Work completed locally doesn’t draw down Perplexity Computer’s cloud credits, meaning users can run recurring or resource-heavy tasks without worrying about hitting usage limits tied to cloud-based processing.
To make the tool actually useful in daily workflows, Perplexity built in connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub. That list covers most of the software stack a typical office or engineering team already relies on, which lowers the barrier to actually adopting the agent rather than treating it as a novelty.
Real-world use cases across industries
Perplexity outlined a handful of concrete scenarios where the tool is meant to save time. In engineering contexts, the agent can review open pull requests inside a connected GitHub project, sort them by status, and flag next steps — it can even spot outdated documentation and draft proposed fixes for review. On the finance side, Portable Computer can be pointed at two years of brokerage summaries, consolidated 1099 forms, and tax returns to trace which recurring holdings are generating the most avoidable fees or tax drag, citing every figure back to the specific file and page it came from — without any document ever reaching a cloud chatbot.
Startups get a different kind of value: the agent can analyze a funnel export locally to figure out where new signups are dropping off between installation and completing their first task, then post the key findings directly into a team’s Slack channel.
Why this matters for the broader AI landscape is worth noting. As local models grow capable enough to handle real analytical work, the pressure on cloud infrastructure — and the privacy tradeoffs that come with sending data off-device — starts to ease for a meaningful slice of everyday tasks. That shift doesn’t eliminate the need for cloud AI, but it does change who controls sensitive information and when it actually leaves a user’s machine.
Perplexity says support for NVIDIA DGX Station is expected to follow soon, which would extend Perplexity Computer Windows compatibility even further across NVIDIA’s hardware lineup. For now, the rollout to GeForce RTX and RTX PRO Workstation owners marks the widest consumer-facing expansion of the company’s local-agent strategy to date.
FAQ
What is Perplexity Portable Computer?
It is a local version of the Perplexity Computer AI agent that executes multistep tasks on Windows PCs with NVIDIA RTX GPUs.
Which hardware is required to run Perplexity Portable Computer on Windows?
NVIDIA GeForce RTX or RTX PRO GPUs with at least 24GB of VRAM are required.
How does Perplexity Portable Computer handle sensitive data?
The local AI analyzes data on the device without sending sensitive information off-device, enhancing data privacy.
What integrations does Perplexity Portable Computer support?
It supports connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
Be the first to comment