Portable Computer: Perplexity and NVIDIA take AI local
Portable Computer pulls Perplexity's AI agent out of the cloud and onto the user's own machine. Perplexity and NVIDIA unveiled the on-device edition of Perplexity Computer, the agent platform introduced earlier in 2026, on August 25, 2026. It is rolling out now to Pro, Max, and Enterprise subscribers as a one-click install in the Perplexity app.
Portable Computer runs the full agent stack locally
Local execution is the entire pitch. The machine itself handles file reads, runs code in an isolated sandbox, preserves session state, and maintains the search index; none of that data leaves the device, and none of it incurs per-token charges. The operating system confines the runtime inside a security sandbox, covering orchestration, subagent models, planning, tool routing, scheduling, and inference. Only real-time information or heavier reasoning escalates to one of more than 15 frontier cloud models, and only with the user's go-ahead.
The intended machines are the NVIDIA DGX Spark, built on the GB10 Grace Blackwell superchip with 128GB of unified memory and 1TB of storage, plus Linux PCs with RTX GPUs and at least 24GB of VRAM. Launch models are Qwen 3.8 27B (4-bit quantized, a 27.6GB download) and PPLX 27B, Perplexity's post-trained variant; NVIDIA's Nemotron 3.5 Lightning (30B) joins the model picker soon. Perplexity's internal benchmarks put Qwen 3.8 27B at 82.6% accuracy, ahead of the Pi and Hermes harnesses. Windows support is planned; Gmail, Outlook, Slack, and GitHub connectors route through the local orchestrator.
Perplexity positions it for data-sensitive, high-volume jobs like repository migration, where cloud token fees balloon. The trade-off is explicit: local models top out near 27B parameters, so users swap some frontier reasoning for privacy, predictable costs, and complete data control.
Why this matters
Portable Computer is the first mainstream subscription agent to bet that local execution can beat cloud-only economics, and the pricing is the tell: zero per-token charges for on-device work. Whether subscribers actually run it on DGX Spark-class hardware will determine how fast rivals follow.
Sources
Portable Computer: Local-First AI
Related Articles
- Nvidia Perplexity investment: the $30B bet turning a chipmaker into an AI power broker
- NemoClaw Deep Agents Blueprint Cuts AI Agent Costs by 90 Percent
- NVIDIA Launches DGX Spark Hardware as Hermes Agent Hits 140,000 GitHub Stars
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.