Perplexity has officially released Portable Computer for Windows PCs powered by NVIDIA GeForce RTX and RTX PRO GPUs, allowing users to run complex agentic AI workflows locally on their own desktops. The update brings unmetered local intelligence and strict data privacy controls to the Windows ecosystem, eliminating cloud token costs for on-device processing.

perplexity local ai agent windows rtx

The official launch of Portable Computer for Windows extends Perplexity's local agent platform beyond its initial deployment on Linux systems and NVIDIA DGX Spark workstations. By packaging the local model, task planner, agent harness, and security sandbox into the native Windows desktop app, Perplexity simplifies local AI deployment for power users and enterprise professionals. Tasks executed strictly on local hardware do not consume monthly Perplexity Computer credits, enabling continuous background processing without recurring token charges.

Perplexity Brings Local Multistep AI Intelligence to Windows PCs

Unlike standard conversational chatbots that simply respond to text prompts, Portable Computer acts as an autonomous agent capable of executing multistep instructions. It plans subtasks, interacts with connected applications, manages local files, and executes complex automated routines in a controlled background environment.

Users can schedule recurring background jobs, such as automated code reviews on GitHub repositories, daily document summaries, or processing raw financial receipts directly into spreadsheets. These capabilities align with broader industry trends toward bringing standalone desktop assistants directly to user environments, similar to how Google has launched a dedicated Gemini desktop app for Windows 11 and 10 to streamline workflow integration.

To ensure security while carrying out multi-step instructions, Portable Computer operates inside a specialized local sandbox environment called the SPACE sandbox. This prevents unauthorized agent actions while granting the AI controlled access to specific system folders and desktop software.

Hardware Requirements and Supported RTX GPUs

Running multi-billion parameter AI models and autonomous agent harnesses directly on consumer hardware requires substantial memory bandwidth and processing capacity. Consequently, Perplexity has set specific hardware prerequisites for Windows machines:

  • GPU Requirement: NVIDIA GeForce RTX or NVIDIA RTX PRO workstation graphics cards.
  • Video Memory (VRAM): A minimum threshold of 24GB of VRAM. Compatible cards include high-end enthusiast desktop GPUs like the GeForce RTX 3090, RTX 4090, and professional RTX PRO series cards.
  • Software Stack: Windows 11 host operating system running the latest Perplexity Windows desktop application from the Microsoft Store.
  • Subscription Tier: Active Perplexity Pro or Max subscription.

The 24GB VRAM requirement acts as a hardware gateway because the system must simultaneously hold the local LLM in memory alongside the agent's context window, orchestration engine, and file parsing structures. NVIDIA's ongoing software support for consumer hardware has made this possible, mirroring recent developments where NVIDIA has released CUDA Toolkit 13.4 to expand GPU acceleration across diverse PC architectures.

Local Execution versus Cloud-Based AI Processing

The primary advantage of Portable Computer is its privacy-first architecture. When analyzing proprietary software codebases, confidential legal files, or sensitive medical records, all model inference, vector indexing, and tool orchestration occur strictly on the user's host machine. Private documents are never uploaded to cloud servers during local tasks.

However, Portable Computer is designed as a hybrid system rather than a permanently offline tool. If a task requires web search capabilities or reasoning beyond the scope of local models, the local agent can escalate the workflow to frontier cloud models such as Claude or GPT-4o. Crucially, the system requires explicit user permission before sending any information off the device.

At launch, Portable Computer on Windows utilizes optimized 27-billion parameter local models such as PPLX 27B, a post-trained version of open-source model architectures specifically tuned for tool calling and reasoning within the Perplexity harness. Perplexity and NVIDIA have also confirmed that support for NVIDIA Nemotron 3.5 Lightning will arrive in an upcoming software update.

Availability and Setup for Windows 11 Users

Windows 11 users with qualified NVIDIA hardware can access Portable Computer immediately by updating to the latest build of the Perplexity Windows desktop application. The setup process eliminates the complex manual configuration steps typically required to manage local AI dependencies, offering a streamlined one-click model download within the settings menu.

The software integrates with native desktop environments and enterprise collaboration tools, featuring built-in connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub. This seamless software integration reflects wider OS updates, such as when Microsoft released Windows 11 preview builds featuring Copilot key remapping to give users direct hardware control over AI applications.

By bringing Portable Computer to Windows PCs, Perplexity and NVIDIA are establishing a practical framework for hybrid AI compute. Power users now gain an efficient method to execute massive, confidential tasks locally on hardware they already own, preserving cloud credits for complex global research.