Perplexity has announced that it is formally bringing its on-device AI agent platform, Portable Computer, to the Windows operating system. This service, originally supporting only NVIDIA DGX Spark and Linux environments, can now run on Windows PCs equipped with NVIDIA GeForce RTX and RTX PRO graphics cards.
By moving the agentic AI workflow to the local machine, users can not only ensure that confidential data never leaks but also harness their own hardware’s computing power to execute complex automation tasks, all without consuming any cloud model credits. Perplexity laid out the details in its announcement that Portable Computer for Windows is here.
Absolute Data Isolation: High Privacy and Zero Credit Consumption
As model capabilities advance, AI agents can already take over many highly confidential desktop tasks. The greatest advantage of Portable Computer lies in its local-first execution logic:
- Sensitive data never leaks: Whether organizing years of tax returns and bank statements or analyzing a startup’s user-churn funnel data, the agent analyzes only locally and precisely cites file sources. The data is never sent to a cloud chatbot.
- Freedom from cloud billing anxiety: Any workflow completed locally does not consume Perplexity Computer’s cloud credit allowance.
- Deep integration with daily productivity tools: The system includes built-in connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub. In a software-engineering scenario, for instance, the agent can automatically review pull requests (PRs) on GitHub and organize the status and next-step suggestions locally.
- On-demand cloud collaboration: When more advanced reasoning or external web search is needed, the system possesses a cloud-escalation mechanism, but it will always seek the user’s authorization before sending instructions to a cloud model.
Ready-to-Use Local Models and a High Hardware Bar
In the past, deploying and running an AI agent locally often required tedious software-environment configuration and model testing. Perplexity greatly simplifies this process through its dedicated app:
- A specially optimized local model: The system includes built-in models such as Qwen 3.8 27B, post-trained specifically for Perplexity Computer and deeply optimized for the NVIDIA RTX architecture.
- Hardware requirements: Portable Computer currently sets a relatively stringent hardware floor, requiring the Windows PC to be fitted with an NVIDIA GeForce RTX or RTX PRO graphics card boasting 24GB or more of video memory (such as the RTX 3090 or higher). The company has also foreshadowed that it will further support the NVIDIA DGX Station in future.
It is worth noting that this architecture currently focuses entirely on the NVIDIA hardware ecosystem and does not support other hardware platforms such as Apple Silicon.
Trading Hardware Investment for AI Compute Freedom
By extending Portable Computer to the Windows platform, Perplexity precisely strikes at the two greatest pain points enterprises and professionals face when adopting AI: data privacy and API cost.
Rather than throwing every task onto the cloud to be handled by frontier models, Portable Computer establishes a hybrid workflow. It hands the tedious “dirty work” that requires repeatedly accessing confidential files to the local RTX graphics card and Qwen model, calling upon a cloud model only when high-level logical reasoning is needed.
This model fundamentally alters the commercial licensing logic of generative AI. Through a one-time investment in a high-end graphics card (at least 24GB of video memory), users trade for unlimited firepower on daily AI-agent tasks. As NVIDIA continues to drive the development of more efficient local models, including Z.ai’s GLM 5.3 Flash and DeepSeek-v4.1 Flash, this tectonic shift from cloud compute toward edge AI will usher in a whole new era of explosive market value for high-end PCs.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
