Perplexity Launches Portable Computer: Local AI Agent Eliminates Cloud Costs and Enhances Privacy

Reviewed byNidhi Govil

14 Sources

Share

Perplexity unveiled Portable Computer, a local AI agent that runs entirely on-device using Nvidia GPUs. The platform eliminates cloud dependency for AI tasks, offering zero token costs for local inference while maintaining data privacy. Available now for Pro and Max subscribers on Linux systems with Nvidia DGX Spark or RTX GPUs with 24GB VRAM, Windows support arrives in September.

Perplexity Portable Computer Brings Agentic AI to Your Desktop

Perplexity has launched Portable Computer, a local AI agent designed to run AI tasks entirely on-device without cloud dependency

1

. Announced in partnership with Nvidia, this local-first agent represents a shift from the company's cloud-based Computer platform introduced in February

2

. The platform runs entirely on user hardware, keeping models, files, and work on the local machine while offering the same interface and features as its cloud counterpart

4

.

Source: Tom's Guide

Source: Tom's Guide

Zero Token Costs and Enhanced Data Privacy

The primary advantage of Perplexity Portable Computer centers on cost savings and data privacy. Local inference avoids per-token API fees entirely, making it possible to handle high-volume tasks without exhausting tokens or incurring extra costs

5

. According to Perplexity, tasks handled locally in an isolated sandbox on your machine reduce the risk of data breaches significantly

1

. Nader Khalil, Nvidia's Director of Developer Technologies, emphasized that the platform lets users lean on AI more broadly without worrying about data security or recurring token costs

4

. Sensitive data never leaves the device without permission, addressing concerns about sending private information like bank statements, contracts, or health records to cloud servers

3

.

Source: VentureBeat

Source: VentureBeat

Hardware Requirements Present a Significant Barrier

Portable Computer demands substantial hardware requirements that may limit accessibility. On Linux systems, the platform requires an Nvidia DGX Spark or other Linux machine with an Nvidia RTX GPU, running Nvidia DGX OS or Ubuntu on ARM or x64 systems

1

. For Windows users, expected to gain access in September, machines must feature an Nvidia RTX GPU with at least 24GB of VRAM

4

. These Nvidia GPUs with 24GB VRAM typically cost at least $1,500, creating a financial barrier for many users

1

. The platform is currently available only to paid Perplexity Pro, Max, Enterprise Pro, and Enterprise Max subscribers

1

.

Source: SiliconANGLE

Source: SiliconANGLE

Model Options and Performance Benchmarks

Portable Computer offers access to small but capable models for agentic workflows. Users can choose between Qwen 3.8 27B, an open-source model recognized for speed, performance, and prowess with coding and complex tasks, or PPLX 27B, a post-trained version refined by Perplexity for enhanced accuracy and efficiency

1

. Support for Nvidia Nemotron 3.5 Lightning, a 30-billion-parameter Mixture-of-Experts model with 3 billion active parameters per token, is coming soon

4

. Benchmarks published by Perplexity show Computer running Qwen 3.8 27B on Nvidia DGX Spark achieved an 82.6% accuracy score using 520,000 tokens on the Local Knowledge Work Bench, exceeding Pi's 77.6% and Hermes' 74.0% across 53 tasks

4

. On PPLX 27B, accuracy reached 85.4% using 678,000 tokens

4

.

Hybrid Approach Maintains Cloud Access When Needed

While emphasizing local inference, Portable Computer maintains the ability to tap into cloud models for complex tasks requiring more powerful frontier models

4

. The agent asks for user permission before sending any data to the cloud, ensuring control over when online resources are accessed

5

. Users can still connect to external services like Google Drive, Gmail, Slack, and GitHub to incorporate cloud-based files and information

1

. This hybrid approach balances the benefits of on-device AI with the flexibility to leverage cloud resources when necessary.

Broader Context of Local AI Movement

Portable Computer arrives amid growing interest in fully local agentic AI setups. Alternative solutions like Eigent, an open-source desktop app built on the CAMEL-AI multi-agent framework, offer similar capabilities for users seeking to keep workflows entirely local

3

. Eigent runs specialized agents for development, search, document processing, and multi-modal tasks, with a Coordinator agent decomposing prompts into subtasks dispatched across independent branches running concurrently

3

. The platform's timing coincides with Nvidia reportedly contemplating a $30 billion investment in Perplexity, though no official confirmation has been provided

2

. As Khalil noted, local AI has reached an inflection point, moving from a hobbyist pursuit toward an indispensable productivity tool

4

.

Today's Top Stories

© 2026 TheOutpost.AI All rights reserved