Perplexity Puts Its Portable Computer Agent on Windows, Running Qwen 3.8 27B Entirely on a 24GB RTX GPU

NVIDIA Blog (corroborated by Tom's Hardware and Unite.AI)·medium signalXBlueskyLinkedInCopy link

Perplexity and NVIDIA announced on September 14 that the local version of the Perplexity Computer agent now runs in the Windows app on GeForce RTX and RTX PRO cards with at least 24GB of VRAM. The on-device model is Qwen 3.8 27B, post-trained for the Computer harness and optimized for RTX, and it reaches other desktop apps through local MCP servers so tool use stays on the machine. Work done locally burns no Perplexity credits and no files leave the device, which is the first mainstream consumer agent to make the local-inference tradeoff explicit in billing.