Nvidia: Perplexity's local agent Portable Computer opens to Windows RTX PCs

英伟达：Perplexity 本地智能体 Portable Computer 已向 Windows RTX PC 开放

Nvidia announced that Perplexity's local agent platform 'Portable Computer' is now open to Windows RTX PC users. As local large model capabilities continue to strengthen, AI agents can handle more tasks directly on personal computers while keeping sensitive information on the device. Portable Computer is the local version of the agent product Perplexity Computer, capable of planning and executing multi-step tasks. It relies on Nvidia GPU acceleration to analyze data locally, integrate information from multiple files, and handle repetitive work. Sensitive data is stored entirely on the local device, and tasks completed on the machine do not consume Perplexity Computer's service quota. Users can also delegate tasks to cloud models for higher-level research and reasoning. Today, Perplexity launched the Portable Computer feature in its Windows client, supporting compatible PCs with Nvidia GeForce RTX GPUs and Nvidia RTX PRO professional workstations, allowing more Windows PC users to use high-performance AI agents. This release is an extension of previously supported Nvidia DGX Spark systems and RTX PCs running Linux. Perplexity integrates local AI with cloud AI within one application, allowing users to process sensitive files on their personal computers and call built-in tools within the Computer, such as a built-in browser and its own SPACE sandbox environment. When a high-level reasoning task is encountered, Portable Computer can automatically identify the need for cloud computing power, requesting user authorization before transmitting data out of the machine. The application simplifies the local model deployment process, with the built-in Qwen 3.8 27B model; the model has been fine-tuned for Perplexity Computer and optimized for Nvidia RTX GPUs. Users do not need to research models or configure complex software environments to enable this agent. The product provides connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub, connecting various files and applications in users' daily workflows to expand use cases. Examples of tasks completed by several agents include: in engineering, reading unmerged code pull requests from a GitHub project connected to the system, categorizing them by status and outlining subsequent work. It can also identify outdated documents and generate modification proposals for review. In finance, it can take two years of broker statements, combined 1099 tax forms, and tax return forms, find assets in continuous holdings that generate large avoidable fees and tax losses, and all data is annotated to corresponding files and pages, with the entire process not uploading files to the chat machine.