Perplexity's local AI agent lands on Windows — bring a 24GB RTX

Perplexity's Portable Computer is the on-device version of its Perplexity Computer agent, and it now runs on Windows PCs with an NVIDIA RTX GPU that has 24GB+ of VRAM, such as an RTX 5090 or an RTX PRO workstation card. Until now it needed a $4,699 DGX Spark or a Linux RTX machine. It ships inside the existing Perplexity Windows app for Pro and Max subscribers.

The default model is PPLX 27B, Perplexity's post-trained version of Qwen 3.8 27B. On DGX Spark it scored 85.4% on the company's 53-task local knowledge-work bench, against 82.6% for stock Qwen. Nemotron 3.5 Lightning (a ~30B MoE with ~3B active params) is marked coming soon. The orchestrator, planner, tool router and task queue all run on the machine, and work done locally uses zero Computer credits.

The Windows build adds two things: local MCP servers, so the agent can drive other desktop apps, and a scheduler that runs recurring jobs while you're away. When a task needs the cloud, it asks before anything leaves the device.

Why it matters: agentic AI is moving from metered cloud tokens to hardware people already own, and a high-end gaming GPU is now enough.