Perplexity’s local AI agent is expanding beyond Linux. In its Sept. 14 announcement, the company said Windows users with Nvidia RTX GPUs can run models and agents on their own PCs, work with local files and connected apps, and use cloud models when needed. According to Nvidia, work completed locally does not consume credits. Portable Computer asks permission before sending content to the cloud. Connected tasks can combine both: analyze signup data locally, then post the findings to Slack. Nvidia previously specified at least 24GB of graphics memory for the Linux RTX version. Perplexity’s Windows announcement does not give a memory minimum.