Perplexity launched Portable Computer on Tuesday, August 25, 2026, introducing an AI agent platform that runs entirely on local user hardware to process complex workflows and integrate documents without relying on cloud servers, according to reports from Tekno, Kompas, and Telset.
Hardware Requirements for Nvidia DGX Spark and RTX GPUs
According to Telset and Kompas, running the system locally demands substantial computing power. On launch, the platform is restricted to high-end hardware, specifically the Nvidia DGX Spark desktop AI computer and Linux systems utilizing Nvidia RTX graphics cards with a minimum of 24GB VRAM.
Hybrid Routing and Cloud Cost Savings
According to Kompas and Telset, local execution means users do not continuously pay token fees or cloud processing credits for routine tasks. However, the system features a hybrid routing mechanism. When a task requires advanced reasoning beyond the local model’s capacity, the platform prompts the user for permission before routing that specific step to a cloud model, charging credits only for the external portion.
Available AI Models and Third-Party Integrations
According to Tekno and Kompas, the initial software release provides access to just two models: Alibaba’s Qwen 3.8 27B and PPLX 27B, which is a post-trained version of Qwen modified by Perplexity. Nvidia Nemotron 3.5 Lightning is scheduled to join the lineup shortly. Despite running locally, the agent connects with external tools like Google Drive, Gmail, GitHub, and Slack to execute commands and share data.
/data/photo/2026/08/26/6a8e6fd5d33ee.jpg "Photo: tekno.kompas.com")
Frequently Asked Questions
What hardware is required to run Perplexity Portable Computer?
According to Telset and Kompas, the platform requires an Nvidia DGX Spark or a Linux computer equipped with an Nvidia RTX GPU.
Does running local AI agents cost cloud processing credits?
According to Telset, local execution does not consume credits. Credits are only billed when the hybrid routing system asks for permission to send a specific step to the cloud.
Which AI models are available at launch?
According to Tekno and Kompas, users can choose between Qwen 3.8 27B and PPLX 27B upon release.
Join the Conversation
What are your thoughts on running resource-heavy AI agents locally versus relying entirely on cloud infrastructure? Share your workflow experiences in the comments below, and subscribe to our newsletter for more updates on hardware-accelerated artificial intelligence.
Related reading