Perplexity and Nvidia Launch Portable Computer: A Fully Local AI Agent
Michael Nuñez | 6:00 am, PT, August 25, 2026 | VentureBeat made with Midjourney
Perplexity is unveiling Portable Computer today, a version of its agentic "Computer" platform that operates entirely on existing user hardware, starting with Nvidia’s DGX Spark desktop supercomputer and Linux machines equipped with RTX GPUs.
This launch represents one of the most significant efforts to shift AI agent workloads from the cloud to local devices. Portable Computer allows tasks to be completed locally with no token costs, ensuring that all data and models remain on the device unless explicitly sent to a cloud-based model.
"We’ve brought the same UI to a fully local app," explained Nate, Perplexity’s vice president of engineering for infrastructure and enterprise during a press briefing. "This includes everything needed for agent harness, inference, and work completion locally."
For Nvidia, this announcement signifies a shift in focus from selling AI data centers to promoting hardware suitable for local AI applications.
"Local AI has reached an inflection point," said Nader, Nvidia’s director of developer technology. "While previously it was mostly hobbyists using tiny quantized models, new open-source models have made local AI more practical."
How Portable Computer Works:
Perplexity Computer orchestrates AI models, files, tools, and web access to perform complex tasks like document reviews, data analysis, report generation, and system integration. Portable Computer replicates this functionality locally:
- Local models
- Agent harness
- Inference engine
- Tools
- App connectors
- Security sandbox
Nate emphasized the ease of use, stating, "Historically setting up a local AI stack was cumbersome. Portable Computer simplifies this process, allowing users to get started quickly."
Demos Highlighting Portable Computer’s Capabilities:
-
Investor Analysis: The system reviewed sensitive financial documents containing 1099s and investment statements locally using a 27-billion parameter Qwen model on a DGX Spark. The results were displayed without incurring cloud credit costs.
-
Hybrid Analysis: Demonstrating the hybrid capabilities, Nate asked the agent to analyze user funnel data locally via CSV and then share the findings on Slack.
This launch signifies a promising step towards more decentralized AI processing, offering potential benefits in terms of privacy, security, and cost efficiency.
