Perplexity's Portable Computer, the locally-running version of its Computer agent, is now accessible inside the Perplexity app for Windows on compatible Nvidia GeForce RTX and RTX PRO GPUs, but users will need an Nvidia GPU with at least 24GB of VRAM to operate it, according to a report published by The New Stack. The Windows release extends Perplexity's reach across three platforms in under three weeks. The launch represents more than a simple software port — it required adapting the model runtime, orchestration layers, security protocols, and hardware integration for each platform while maintaining a consistent user experience.
On Windows, Portable Computer bundles PPLX 27B — Perplexity's post-trained model — and Qwen 3.8 27B, both tuned for RTX GPUs, along with an integrated browser, tool calling capabilities, and Perplexity's proprietary SPACE sandbox. The software ships with connectors for Microsoft Outlook, OneDrive, and Word, plus Google Drive, Gmail, Slack, and GitHub. The platform launched first on Linux and Nvidia DGX Spark on August 25, followed a week later by hybrid compute for Apple silicon, which divides tasks between local and cloud models on Macs. Portable Computer is available with Perplexity Pro at $20 monthly and Max at $200 monthly, spanning individual and enterprise plans, with Nvidia DGX Station support planned for later.
According to The New Stack, Portable Computer operates in a different category from LM Studio or Ollama, which focus on making local model operation as simple as possible but fall well short of granting a model autonomy over multistep work. The report notes that DeepSeek's recent hiring push of roughly 150 new positions, nearly all centered on agent infrastructure rather than the model itself, suggests how much engineering exists between a capable model and a capable agent. When the agent determines that a task requires more reasoning capacity than the local model can provide, it can escalate to Perplexity's cloud models, and Nvidia states the agent identifies when cloud support would assist and requests user permission before transmitting any data off the machine.
The hybrid architecture matters because Perplexity isn't claiming that a 27-billion-parameter model running on a desktop GPU can manage everything. For organizations working with sensitive or regulated data, that division can prove decisive — a local agent can process source code or financial records without uploading them to a hosted model for basic operations. There's a financial dimension as well, since tasks finished locally don't consume Perplexity Computer credits. The agent can reach external services because it isn't air-gapped, meaning locally completed tasks can handle files without routing documents to a cloud model. But once an agent accesses both local files and remote APIs on the same machine, determining which resources it actually needs and where to locate them becomes more complex. The report identifies the real challenge as moving local agents from developer passion projects to enterprise-ready tools, noting that by integrating this into Windows, it immediately reaches the scale of users required to make that transition happen. The 24GB VRAM floor remains the most significant barrier to broader adoption, limiting the technology to high-end workstations and excluding the majority of consumer and business PCs from running these agents independently. Organizations will need to weigh whether the control and cost savings of local processing justify the hardware investment, or whether cloud-based agents remain the more practical path for most workflows.