By dev.ua
Publication Date: 2026-08-27 00:00:00
Perplexity and Nvidia have unveiled a joint local AI platform called Portable Computer. It is an optimized version of Perplexity’s Computer platform that allows you to perform complex tasks directly on your own hardware — without resorting to cloud servers and without spending on tokens.
During the press conference, Perplexity representative Nate Kapp noted that the application will retain the familiar interface of the original platform. For particularly complex tasks, the system may offer to involve more powerful cloud models, but it does so only with the user’s consent, reports TechSpot.
According to Nvidia’s Chief Technology Officer for Developers Nader Khalil, local AI has reached a tipping point and is transforming from an enthusiast’s fascination to an everyday productivity tool that ensures complete data security.
Technical specifications and architecture:
the application consists of an agent runtime, an orchestrator, and local models;
Qwen 3.8 and PPLX are available (both scale to 27 billion parameters). Support for Nvidia’s Nemotron 3.5 Lightning, a free 30 billion parameter MoE model (with 3 billion active parameters per token), will be added soon.
Requires a Linux PC with an Nvidia RTX graphics card (minimum 24 GB VRAM) or a dedicated DGX Spark micro-PC (based on the Grace Blackwell GB10 chip with 128 GB of combined memory). Windows support is promised to be added in September with a similar requirement of 24 GB of video memory.
Test results
On the…
