Perplexity launches Portable Computer, a local-first AI agent for NVIDIA DGX Spark

Perplexity has launched Portable Computer, a version of its Perplexity Computer agent that runs entirely on local hardware instead of the cloud, built jointly with NVIDIA for the DGX Spark desktop AI machine.

What's new

"Today we're launching Portable Computer, a version of Perplexity Computer that runs entirely on a local machine," the company said, adding it built the product with NVIDIA "so users can own the stack" on the DGX Spark, NVIDIA's desktop-sized AI computer, with NVIDIA RTX GPU PC support coming soon.

The entire agent stack runs on-device: orchestrator, planner, tool router, scheduler, durable task queue, and local search index. It ships with a choice of Qwen 3.8 27B or PPLX 27B (a post-trained version of the Qwen model) as launch models, with NVIDIA's 30B open Nemotron 3.5 Lightning model coming soon to the picker. Dictation runs locally too, using the NVIDIA Nemotron 3.5 ASR model, so a user can talk through a task or dictate over a private codebase without audio leaving the machine.

Private data stays on the device, and on-device work doesn't consume credits. When a task needs current information, browser use, connected apps (Google Drive, Gmail, Slack, GitHub), or one of Perplexity's 15+ frontier cloud models, the local orchestrator can escalate — but only after asking the user for permission. Security is designed to match the cloud version of Perplexity Computer, with code and tool execution running in isolated sandboxes.

Portable Computer is available now to Perplexity Pro and Max subscribers on the DGX Spark, which is built on NVIDIA's Grace Blackwell GB10 platform (a 20-core Arm CPU paired with a GPU and 128GB of unified memory). The first release supports Linux, with Windows support coming soon, and installs via a one-click setup in the Perplexity app.

Context

Portable Computer is a local counterpart to Perplexity Computer, the company's cloud-based agent for multi-step tasks like building web applications. Pairing it with NVIDIA's DGX Spark reflects a broader industry push toward capable on-device inference: as local models close the gap with frontier cloud models, running sensitive workflows entirely on a user's own machine becomes viable without sacrificing much capability. NVIDIA covered the launch on its own Local AI blog, and Perplexity published a companion research post describing Portable Computer's development and its benchmark performance on accuracy, speed, and credit efficiency.

Why it matters

The no-credit-cost, fully local design targets a real pain point for high-volume AI users: workloads that are either too sensitive to send to the cloud (confidential codebases, deal documents) or too frequent to run economically against metered cloud credits. By defaulting to local execution and escalating to the cloud only on request, Perplexity is betting that on-device compute — not just bigger cloud models — will define the next phase of agentic AI usage. It also deepens Perplexity's dependency on NVIDIA hardware as a differentiator, tying a software product directly to a specific desktop AI computer rather than shipping a hardware-agnostic local mode, a bet that pays off if DGX Spark adoption grows but limits reach in the near term to Pro/Max subscribers who own the device.