Perplexity has extended its Portable Computer offering to Windows systems running compatible NVIDIA GeForce RTX and RTX PRO graphics cards. The application packages together model runtime, orchestration, security controls, and hardware integration for executing AI agents locally. The significant hardware requirement is a GPU with minimum 24GB of VRAM.
The Windows release marks Perplexity's third platform launch in under three weeks. Portable Computer first appeared on Linux and NVIDIA DGX Spark on August 25, followed by hybrid compute support for Apple silicon one week later, which distributes processing between local and cloud models on Macs. Bringing the application to Windows required more than a straightforward port; Perplexity adapted the model runtime, orchestration mechanisms, security architecture, and hardware integration for each platform while maintaining consistency in user experience.
Orchestration beyond the model
On Windows, Portable Computer supports PPLX 27B—Perplexity's post-trained model—and Qwen 3.8 27B, both tuned for RTX GPUs. The package includes an integrated browser, tool calling capabilities, and Perplexity's proprietary SPACE sandbox. This positioning differs from tools like LM Studio or Ollama, which prioritize ease of local model deployment but do not grant models autonomy over complex, multi-step operations. DeepSeek's recent hiring initiative brought on roughly 150 new employees, with nearly all positions centered on agent infrastructure rather than model development, underscoring the engineering complexity between a functional model and a functional agent.
Connectors blur local boundaries
Perplexity includes connectors for Microsoft Outlook, OneDrive, and Word, as well as Google Drive, Gmail, Slack, and GitHub. This roster reveals what "local" execution actually entails in practice. The agent can interface with external services because it operates without air-gap isolation. Tasks processed locally can handle files without transmitting documents to cloud-based models. Once an agent gains access to both local files and remote APIs on the same device, determining which resources it requires and their locations becomes more complex.
Hybrid cloud as fallback
Perplexity acknowledges that a 27-billion-parameter model on a desktop GPU cannot address all scenarios, which justifies the hybrid design. When the agent recognizes that a task exceeds the local model's reasoning capacity, it can escalate to Perplexity's cloud infrastructure. According to NVIDIA, the agent detects when cloud assistance would be beneficial and requests user authorization before transferring any data off the machine.
For enterprises managing sensitive or regulated information, this separation offers substantial advantages. A local agent can process source code or financial records without uploading them to hosted models for routine operations. Cost considerations also apply, since locally completed tasks do not consume Perplexity Computer credits.
High VRAM floor limits reach
Portable Computer is available through Perplexity Pro ($20/month) and Max ($200/month) subscriptions, offered across individual and enterprise plans, with NVIDIA DGX Station support arriving at a later date. The principal obstacle involves transitioning local agents from experimental developer projects into production-grade enterprise solutions. By integrating this capability into Windows, Perplexity immediately reaches the user volume necessary to drive that transition forward.