🛠️Perplexity Runs Its Full Agent Locally on Nvidia DGX Spark
TL;DR
Perplexity's Portable Computer runs the entire agent harness, inference and file access on hardware you already own. Local work burns zero billing credits, and the agent asks before sending any step to a cloud model.
Perplexity's Portable Computer runs the entire agent harness, inference and file access on hardware you already own. Local work burns zero billing credits, and the agent asks before sending any step to a cloud model. Nvidia co-built it, which says something about where it thinks inference is heading.
Key Points
Shipped August 25, 2026 for Nvidia DGX Spark and Linux machines with RTX GPUs
Available to Pro, Max, Enterprise Pro and Enterprise Max subscribers; Windows support lands in September
Work finished on-device consumes no billing credits; every task starts local by default
The system requests permission before escalating an individual step to a cloud frontier model
Perplexity VP of infrastructure engineering Nate: 'This incorporates the entirety of the agent harness and inference and everything needed to do work locally'
Why It Matters
Agent token spend and data movement are the two things enterprises cannot govern today. Running the harness on-prem attacks both at once, and it is the first credible answer to per-seat agent billing.
Quick Facts
Frequently Asked Questions
Why does this matter?
Agent token spend and data movement are the two things enterprises cannot govern today. Running the harness on-prem attacks both at once, and it is the first credible answer to per-seat agent billing.
What happened?
Perplexity's Portable Computer runs the entire agent harness, inference and file access on hardware you already own. Local work burns zero billing credits, and the agent asks before sending any step to a cloud model.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,317 builders reading daily.