Perplexity introduces Portable Computer, a local-first AI agent designed to run on user hardware with Nvidia’s DGX Spark, enhancing privacy and reducing token costs for sensitive material handling.
Perplexity has launched Portable Computer, a local-first version of its AI agent that is designed to run on a user’s own hardware rather than in the company’s cloud. The move is aimed at people handling sensitive material, as well as subscribers who want to avoid token-based usage costs. According to Perplexity and NVIDIA, the system can operate on the new DGX Spark desktop machine and, with restrictions, on RTX-equipped PCs that have enough memory.
The software changes the way Perplexity’s agent works. Instead of sending every request to remote servers, core parts of the system, including planning, scheduling, tool routing and the search index, now run locally. Only more demanding tasks are handed off to cloud models, and the user must approve that step. VentureBeat reported that the default is on-device processing, with cloud use reserved for more complex reasoning.
NVIDIA says the DGX Spark was built for this sort of workload. The compact system uses the GB10 Grace Blackwell Superchip and is meant to support autonomous agents and large models locally, with a focus on always-on desktop use. Perplexity’s current local stack includes Qwen 3.8 27B with 3-bit quantisation and its own PPLX 27B model, while NVIDIA’s Nemotron 3.5 Lightning and local speech recognition are due to follow.
There are still clear limits. Perplexity and NVIDIA say users need at least 24GB of VRAM on an RTX card, which excludes much consumer hardware, and the DGX Spark remains an expensive reference option. The local models are also smaller than the cloud systems, so the cloud fallback is part of the design rather than an afterthought. Even so, the launch is a sign of where the market is heading: more AI vendors are now selling on-device execution as a privacy feature and a way to bring agentic software into company networks without relying entirely on external servers.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
