Perplexity has added Portable Computer to its Windows app for compatible Nvidia GeForce RTX PCs and Nvidia RTX PRO Workstations. The local agent requires an RTX GPU with 24GB or more of VRAM, while support for Nvidia DGX Station is expected to follow soon.
Portable Computer is a local version of Perplexity Computer that plans and carries out multistep tasks. According to Nvidia, it uses local models accelerated by Nvidia GPUs to analyze data, combine information across files and manage recurring work while keeping sensitive information on the device; tasks completed locally do not consume Perplexity Computer credits.
The system also connects local processing with cloud models for work requiring more advanced research or reasoning. Nvidia says Portable Computer can identify when cloud support is needed and ask the user for permission before sending information off-device. Within the Perplexity app, users can also access Computer’s built-in browser and proprietary SPACE sandbox.
The app is designed to reduce the setup normally required to run local AI. Nvidia cited Qwen 3.8 27B as an example of a local model post-trained to work with Perplexity Computer and optimized for Nvidia RTX GPUs, allowing users to begin using the agent without independently researching models or configuring a local software stack.
Connectors are available for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub. Nvidia said possible workflows include reviewing and organizing open pull requests, identifying outdated documentation and proposing updates; tracing fees and tax drag across brokerage summaries, consolidated 1099s and tax returns with citations to exact files and pages; and analyzing a startup’s funnel export locally before posting findings to Slack.
The Windows release extends Portable Computer beyond its existing support for Nvidia DGX Spark systems and RTX PCs running Linux. Nvidia positioned the expansion as a way for more Windows users to run agentic tasks locally while retaining the option to escalate selected work to cloud models with permission.