Perplexity Introduces Portable Computer, a Local-First Version of Its AI Agent
Perplexity has introduced Portable Computer, a locally-run version of its Perplexity Computer agent that performs AI-powered analysis, file synthesis, and complex workflows entirely on a user’s own machine, escalating to the cloud only when a task requires it.
Details
- What it is: A local-first variant of Perplexity Computer, built to keep sensitive data — confidential documents, private codebases — on-device by default
- Hardware partnership: Built together with NVIDIA specifically for the NVIDIA DGX Spark, a desktop-sized AI computer built on the Grace Blackwell GB10 platform (20-core Arm CPU, NVIDIA GPU, 128GB unified memory); future support planned for NVIDIA RTX GPU PCs
- On-device models and stack: Runs Qwen 3.8 27B or PPLX 27B (a post-trained Qwen variant) locally, with a local search index, orchestrator, and scheduler; local processing carries no per-credit charges
- Local dictation: Voice input uses NVIDIA’s Nemotron 3.5 ASR model running on-device
- Cloud escalation, on request: Users can selectively authorize escalation to the cloud for live web access or heavier reasoning; the system asks permission before any content leaves the device
- Architecture (per Perplexity’s companion technical post): Described as a harness and model co-designed for local knowledge work — running on-device by default and reaching out to remote capabilities only on demand
- Availability: Rolling out to Perplexity Pro and Max subscribers, launching first on Linux via one-click setup inside the Perplexity app, with Windows support “coming soon”
What happened next
Portable Computer positions Perplexity to compete for privacy-conscious professional users — those handling confidential material who have been reluctant to send data to cloud-based AI agents. The NVIDIA DGX Spark partnership also signals a broader industry push toward capable on-device AI hardware; expect Perplexity to expand platform support (Windows, RTX GPU PCs) and possibly extend the local-first approach to other parts of its product line following this launch.