Perplexity's AI Agent Now Runs on Your Own Windows PC, If You Own the Right Graphics Card
Portable Computer arrived on Windows for RTX cards with 24GB of memory or more. It runs a 27 billion parameter model locally, connects to Outlook, Gmail, Slack and Drive, and asks permission before sending anything to the cloud.
Perplexity brought its Portable Computer agent to Windows on Monday, in partnership with Nvidia. It is the local version of Perplexity’s Computer product: an assistant that carries out multi step tasks, except that the thinking happens on the machine in front of you rather than in a data centre. Until now it ran on Linux and Nvidia’s DGX Spark boxes. The Windows requirement is specific: a GeForce RTX or RTX PRO card with at least 24GB of video memory, which in practice means a high end desktop card rather than the laptop most people own.
The model doing the work is Qwen 3.8 27B, tuned for RTX hardware, and you can swap in PPLX 27B, Perplexity’s own post trained variant, with Nvidia’s Nemotron 3.5 Lightning promised later. The useful part is the connectors: Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub, so the agent can work on your actual mail and files rather than on things you paste into a box. When a task genuinely needs more reasoning than the local model has, it can escalate to a cloud model, but only after asking. Local work does not consume the credits that cloud usage does, which means the running cost of an hour of local agent time is your electricity bill. Access comes with a Perplexity Pro subscription at 20 dollars a month or Max at 200.
What is behind this
Two things are converging. Open weight models in the 20 to 30 billion parameter range have quietly become good enough for ordinary work, and consumer graphics cards now have enough memory to hold one. Parameters are roughly the adjustable numbers a model learned during training, and the memory a card has determines how large a model it can keep loaded, which is why the 24GB line exists. The pitch is privacy, and here it is more than a slogan, because an agent with access to your inbox is exactly the case where “the data never leaves this machine” is worth something concrete. The thing to keep in perspective is capability. A 27 billion parameter model on your desk is competent, not frontier, and the escalate to cloud button exists because Perplexity knows that.
What this means for you: Check your graphics card before you get excited, because the 24GB requirement rules out most laptops and a good share of gaming desktops. If you clear the bar and already pay for Perplexity, this is the most practical local agent setup shipped so far, and it is worth trying on the boring jobs first: sorting a week of mail, pulling figures out of your own documents, drafting replies. If you do not clear the bar, the news is still useful as a direction. What needed a data centre two years ago now needs a good graphics card, and models keep getting smaller for the same quality. For anyone who handles client data, patient data or anything under a confidentiality duty, this is the category worth watching, because it is the only one where the privacy question has a clean answer.
Sources
Source: https://blogs.nvidia.com/blog/local-ai-perplexity-windows-pcs/
Real People Are Reading Real ChatGPT Conversations, and Most Users Have No Idea
404 Media obtained internal documents describing Project Lily, OpenAI's programme of hiring hundreds of contractors to read genuine user prompts and grade the answers. Personal details slip through the filters, and the setting that controls it is on by default.