Perplexity has released a new version of its Personal Computer agent, called Portable Computer, which runs on local AI models instead of relying on cloud-based processing. The tool, debuting Tuesday, offers the same standard AI commands and agentic features as the February release, but keeps models, files, and work on the user鈥檚 machine. Users decide if and when the AI connects to the internet, though it can still integrate with external services like Google Drive, Gmail, Slack, and GitHub.
The shift to local processing addresses three key drawbacks of cloud AI, according to the company. First, local models can respond faster than online counterparts. Second, tasks run in an isolated sandbox on the device, lowering the risk of data breaches. Third, local operation does not consume billing credits or tokens, while any part of a task that does go online will eat into those resources.
Perplexity stated in a technical blog post that Portable Computer is a local-first agent, meaning sensitive data never leaves the device without permission and local models carry no inference fee, making the system private and cost-effective by design. For now, the tool is available only on Linux systems, with Windows support promised for September. The hardware requirements are steep, however, which limits who can use it.
On Linux, Portable Computer requires an Nvidia DGX Spark or another Linux machine with an Nvidia RTX GPU, running Nvidia DGX OS or Ubuntu on either ARM or x64 architecture. On Windows, the PC must have an Nvidia RTX GPU with at least 24GB of VRAM. That is a significant barrier, as such graphics cards typically cost at least $1,500, and many desktops, including those with lower-end RTX GPUs, will not meet the spec.
Users have two model options at launch. One is Qwen 3.8 27B, an open-source model praised for speed, performance, coding, research, and complex agentic tasks. The other is PPLX 27B, a post-trained version that Perplexity has refined for greater accuracy and efficiency. Nvidia鈥檚 Nemotron 3.5 Lightning, designed for high-volume, long-running tasks, is expected to be added soon, and users will be able to switch between models depending on the assignment.
The release represents a new path for local AI, though the demanding hardware requirements may keep adoption limited for now. Perplexity鈥檚 move suggests that as local models improve, more users could seek the performance, security, and cost benefits of keeping AI off the cloud. Whether the requirements will ease over time remains to be seen, but the option is now available to those with the right equipment.
More AI news from TechManNews.
