Perplexity and Nvidia ship Portable Computer, a local AI agent with no token billing

Perplexity's Portable Computer bundles local models, agent harness, inference engine, connectors, and an OS-level sandbox into one app for Nvidia DGX Spark and RTX GPUs with 24GB+ VRAM, running Qwen 3.8 27B or Perplexity's post-trained PPLX 27B at zero credit cost. On Perplexity's own Local Knowledge Work Bench the co-designed harness scored 82.6% against 77.6% for Pi and 74.0% for Hermes on the identical model, with 70% fewer tokens on BrowseComp. Any developer with a 3090-class card or newer can now run hours-long agent workloads against Gmail, GitHub, Drive, and Slack without an API meter, escalating to a cloud model only on permission. The benchmarks are self-reported, and Windows support lands in September.

Source: venturebeat.com ↗