Portable Computer launched today as a version of Perplexity’s “Computer” platform that runs on hardware users already own, beginning with Nvidia’s DGX Spark desktop supercomputer and Linux machines equipped with Nvidia RTX GPUs.
Local AI moves off the cloud with Portable Computer
The product arrives after a close partnership with Nvidia, marking one of the most aggressive pushes to shift AI agent workloads from remote servers to local AI devices. All model files, user data and the work itself stay on the machine, so no billing credits are consumed for tasks that run locally.
“We’ve basically brought the exact same UI to a fully local app,” said Nate, Perplexity’s vice president of engineering for infrastructure and enterprise, during a press briefing. He added that the release “incorporates the entirety of the agent harness and inference and everything needed to do work locally.”
Nvidia’s director of developer technology, Nader, said the move reflects a belief that local AI has shifted from hobbyist curiosity to a practical tool. He noted that earlier “quantized models … were super tiny,” but recent open‑source models are “super useful.”
How the bundled stack works
Perplexity’s original Computer platform coordinates models, files, tools and web access to complete multi‑step tasks such as reviewing document folders, analyzing data and generating reports. It packages the same experience into a single app, including the local model, agent harness, inference engine, tools, connectors and a security sandbox.
Related: Nvidia simplifies AI with basic math approach
Most local AI setups require users to download model weights, launch an inference server, link tools and tune performance. “Historically it’s just been really painful to bring up the local AI stack,” Nate said. “With Portable Computer, we really focused on just making this a really straightforward experience where you can get up and running very quickly.”
In a demo, the system acted as a retail investor reviewing a folder of 1099s and investment documents. Running a 27‑billion‑parameter Qwen model at full GPU utilization on a DGX Spark, the agent flagged unnecessary fees. The usual cloud‑credit counter stayed at zero, because the work never left the device.
The second demo showed hybrid capability: after analyzing a CSV of user‑funnel data locally, the agent pushed the results to a Slack channel using Perplexity’s connector ecosystem.
At launch, users may set up the Qwen model or a Perplexity‑trained PPLX model, with Nvidia’s Nemotron 3.5 Lightning promised later. The offering is available today for Pro, Max, Enterprise Pro and Enterprise Max subscribers on Linux.
From a practical standpoint, this means organizations handling sensitive financial or legal documents can keep those files on‑premise while still leveraging sophisticated AI assistance.
Related: Serval’s Catalyst Deploys Agents to Fix IT Issues
Local AI is now truly on‑premise.
The zero‑token cost eliminates a recurring expense that many enterprises face when agents run in the cloud, and it sidesteps data‑movement concerns that regulators often flag.
Performance claims and token economics
The internal “Local Knowledge Work Bench” benchmark, covering 53 tasks, gave its Computer running Qwen 3.8 27B on a DGX Spark a score of 82.6 %, compared with 77.6 % for the open‑source Pi harness and 74.0 % for Hermes using the same model. The post‑trained version reached 85.4 %.
On the BrowseComp web‑research benchmark, Computer hit 66.7 % accuracy, while Pi managed 50.2 % and Hermes 43.9 %. It also used 51 % less wall‑time and 70 % fewer tokens than Pi. Multimodal document understanding saw Computer at 65.1 % versus Hermes 34.6 % and Pi 13.9 %.
