Perplexity has launched Portable Computer for Windows, extending its local-first artificial intelligence agent to compatible Nvidia RTX PCs. The software is designed to carry out AI processing and tool execution directly on a user’s computer, while requiring permission before task data is sent to cloud-based models.
The Windows release moves Portable Computer from a previously announced September rollout to an available product. However, its hardware requirements mean the technology is currently aimed primarily at high-performance PCs and workstations rather than typical office laptops.
Portable Computer Arrives on Windows RTX Systems
Portable Computer is available through the Perplexity Windows application for supported systems equipped with an Nvidia GeForce RTX or RTX PRO graphics processor and at least 24GB of video memory, or VRAM.
The local model currently available for Windows is PPLX 27B. Access is offered to Perplexity Pro and Max subscribers, including eligible individual and enterprise customers.
Nvidia announced the Windows release on 14 September, confirming that Portable Computer could run on compatible RTX PCs and professional workstations.
Perplexity’s product documentation also specifies a minimum of 24GB VRAM for local inference. Qwen 3.8 27B is not currently available on Windows RTX machines, while Nvidia Nemotron 3.5 Lightning is listed as coming soon.
Hardware Requirements Limit Wider Deployment
The 24GB VRAM threshold is significant because many mainstream business PCs do not carry graphics hardware with that level of dedicated memory.
For New Zealand businesses considering local AI deployment, particularly organisations seeking greater control over commercially sensitive information, hardware capability will therefore be one of the first considerations.
Companies assessing Portable Computer across an existing device fleet will need to check individual GPU specifications rather than assuming that any RTX-equipped computer can run the local model.
AI Tasks Can Be Processed Locally
Portable Computer can work with documents, spreadsheets, PDFs, source code, images and other files stored in locations approved by the user.
According to Perplexity, its model, orchestrator, scheduler and execution environment can operate locally. Code and tool execution take place in isolated environments with controlled access to files and connected applications.
The local orchestrator can also work with services including Gmail, Outlook, Slack and GitHub. Desktop applications can connect through local Model Context Protocol servers.
Work completed entirely by the local model does not use Perplexity Computer credits.
Cloud Processing Requires User Approval
Some tasks may still require external services. Portable Computer can use Perplexity Search or access more than 15 cloud models when a task requires current web information or more advanced reasoning capabilities.
Perplexity says users must approve the transfer before information from their device is sent to a cloud model.
This local-first approach can reduce reliance on cloud processing, but it does not remove questions around permissions, security and data governance.
IT administrators still need to understand what information individual connectors can access, how activity is logged and whether the agent’s local execution boundaries comply with existing organisational security policies.
The issue is particularly important when AI agents are allowed to interact with local files. A separate vulnerability disclosed earlier in 2026 affected Perplexity’s Comet browser and involved potential local-file exposure through malicious calendar content.
That vulnerability was unrelated to Portable Computer, but it highlights the importance of carefully managing file permissions and defining boundaries for autonomous AI tools.
What IT Teams Should Check Before Deployment
Portable Computer is available to Pro and Max customers across individual and enterprise subscriptions.
Perplexity’s enterprise products offer features including single sign-on, SCIM provisioning, audit logs and data-retention controls. However, the Portable Computer documentation does not detail how every enterprise security feature applies specifically to local agent activity.
Before beginning a workplace pilot, organisations should assess connector permissions, access to local folders, cloud-escalation logging, sandbox protections, model-update controls and restrictions governing tool execution.
Hardware compatibility should also form part of the initial assessment. With at least 24GB of GPU memory required, organisations may need dedicated AI-capable workstations rather than relying on their existing general-purpose business computers.
Controlled Pilots Could Help Assess Business Use
Initial testing should ideally use non-production accounts, restricted folders and narrowly configured connectors. This can help IT teams understand how Portable Computer interacts with company information before access is expanded.
For New Zealand organisations, the arrival of more capable on-device AI could be relevant where data handling, latency or dependence on cloud services are important considerations. However, local processing alone does not remove the need for established cybersecurity, identity-management and information-governance practices.
Portable Computer is now available on compatible Windows systems, but its suitability for wider workplace adoption will ultimately depend on whether its hardware requirements, security controls and data-handling model align with each organisation’s technology environment.
Alexander Donovan writes for News Collective, covering news, politics, business, technology, sport, entertainment, and lifestyle. He focuses on clear, reliable reporting and useful information, helping readers stay informed about current events, emerging trends, and stories that matter.