The agent runs through the Perplexity app and uses a locally deployed model such as Qwen 3.8 27B, which has been post-trained for Perplexity Computer and optimized for NVIDIA RTX hardware. Perplexity packages the model alongside its Computer tools, including a built-in browser and proprietary SPACE sandbox, so users do not have to configure a separate local AI software stack.
The system can also move between local and cloud processing. If Portable Computer determines that a task requires more advanced reasoning, it can ask the user for permission before sending information off the device.
“As local models become more capable, AI agents can handle more work directly on a PC while keeping sensitive information on the device,” NVIDIA’s Gerardo Delgado said.
That local-first design is particularly relevant for workflows involving confidential files. NVIDIA highlighted a finance example in which Portable Computer can review two years of brokerage summaries, consolidated 1099s and tax returns, identify recurring holdings associated with fees and tax drag, and cite the exact files and pages supporting its findings without sending the documents to a hosted chatbot.
The agent also connects with Microsoft Outlook, OneDrive and Word, along with Google Drive, Gmail, Slack and GitHub.
Those integrations give Portable Computer access to broader workplace tasks. In engineering, it can review open pull requests in GitHub, group them by status, identify next steps and flag documentation that may need updating. For startup teams, it can examine a funnel export locally, identify where users are dropping off and send its findings to Slack.
“Users can put the agent to work without having to research models or configure the complex software stack typically required to run local AI,” Delgado said.
The hardware requirements put limits on who can use the Windows release. Portable Computer requires a GeForce RTX or RTX PRO GPU with at least 24GB of VRAM, meaning the feature is aimed at more powerful PCs and workstations rather than the full Windows installed base.
Support for NVIDIA DGX Station is expected to follow.
For Perplexity, Portable Computer creates a hybrid model for agentic work. Tasks that fit within the capabilities of the local model can remain entirely on the user’s machine, while more demanding research or reasoning can still be routed to cloud models with permission.
That setup gives users another option for workflows where sending files to a hosted AI service may be undesirable. Rather than making cloud processing the default, Perplexity is putting local execution first and treating remote models as an escalation path when the task requires more capability.
This analysis is based on reporting from Nvidia.
Image courtesy of Nvidia.
This article was generated with AI assistance and reviewed for accuracy and quality.
