PERPLEXITY LAUNCHES FULLY LOCAL AI AGENTS WITH $NVDA Perplexity is launching Portable Computer, a local version of its agentic Computer platform built with NVIDIA that can run models, tools, files and multi-step AI workflows entirely on-device. The key difference: local work uses zero Perplexity credits and carries essentially zero marginal token cost. Tasks start locally by default, and the system asks permission before sending any step to a frontier cloud model. Hardware support starts with NVIDIA DGX Spark and Linux PCs with RTX GPUs carrying at least 24GB of VRAM, roughly an RTX 3090 or newer. Windows support is planned for September. At launch, users can run Qwen 3.8 27B or Perplexity’s post-trained PPLX 27B locally, with NVIDIA Nemotron 3.5 Lightning coming next. The platform bundles the full local stack: • Model inference • Agent harness • Tools and connectors • Security sandbox • Local file access • Gmail, Google Drive, GitHub and Slack integrations Perplexity says PPLX 27B scored 85.4% on its internal Local Knowledge Work Bench, versus 82.6% for Qwen 3.8 27B running through its own Computer harness. On BrowseComp, its local system scored 66.7%, while using 70% fewer tokens and 51% less time than the Pi harness. The hybrid setup is also notable. On Terminal Bench 2.1: • Fully local Qwen: 59.6% at near-zero marginal inference cost • Local + Claude Opus 5 advisor: 73.0% at ~$0.415/task • Claude Opus 5 alone: 82.4% at ~$0.65/task Before any cloud escalation, Perplexity says the system scans outgoing context for PII and shows users what data would leave the device. The remote model only returns text guidance and never directly accesses local files or tools. NVIDIA also says the hardware can scale beyond a single machine. Two DGX Sparks can run larger frontier-class open models, while four can handle models such as GLM 5.2 or Nemotron Ultra.