Perplexity Portable Computer Delivers Local AI Agents to NVIDIA RTX Windows PCs

Perplexity Portable Computer is a local version of its Perplexity Computer agent designed to run multistep tasks entirely on supported NVIDIA RTX hardware.

The introduction of Perplexity's Portable Computer on Windows RTX PCs represents an important advancement in the field of local AI agents. This system allows for the execution of complex, multistep tasks directly on the user's hardware. By leveraging the power of NVIDIA RTX GPUs with at least 24GB of VRAM, the agent can perform without sending data to remote servers. This approach addresses key concerns in the AI industry regarding privacy and operational costs. Users with Pro or Max subscriptions can now access this feature through the Perplexity Windows app available on the Microsoft Store for Windows 10 and Windows 11. The local models such as Qwen 3.8 27B, which has been post-trained for Perplexity Computer and optimized for RTX GPUs, enable this on-device operation. Similarly, the PPLX 27B model is available for local use. Support for NVIDIA Nemotron 3.5 Lightning is anticipated in the near future. This development builds on existing support for NVIDIA DGX Spark systems and Linux based RTX PCs, with DGX Station support expected to follow shortly.

What led to the expansion of Perplexity Portable Computer to Windows platforms?

Perplexity has been working on agentic AI systems that can handle multiple steps in tasks autonomously. The Portable Computer is designed to bring this capability to local environments. Previously, the focus was on specialized systems like the DGX Spark. The move to Windows PCs opens the technology to mainstream high-end users who own GeForce RTX and RTX PRO GPUs. This is particularly relevant for professionals and enthusiasts who require high performance computing on their desktops. The 24GB VRAM requirement ensures that the hardware is capable of running the large models necessary for agentic behavior. By making it available on Windows, Perplexity is responding to the growing demand for local AI solutions that do not depend on constant internet connectivity or cloud services. The integration with the Microsoft Store app simplifies installation and updates for Windows users.

The emphasis on local execution means that sensitive data remains on the device. This is a key feature for users in industries where data security is paramount. Tasks that do not require external data can be completed without any cloud interaction. When advanced reasoning or web data is needed, the system requires explicit user permission before escalating to the cloud. This permission-based approach gives users full control over their information. It also means that local work does not use up any Perplexity Computer credits, making it cost effective for frequent use. The combination of these features positions the Portable Computer as a tool for both personal and enterprise applications where privacy and budget are concerns.

What are the specific new features in the Windows version of the agent?

The Windows release brings the full suite of agent capabilities to RTX equipped PCs. The agent can handle multistep tasks such as research, planning, and execution all locally. The models run on the GPU, taking advantage of the hardware acceleration provided by NVIDIA's architecture. This local processing power allows for faster response times compared to cloud based alternatives in some cases. The requirement for a Pro or Max subscription ensures that only paying users have access, which helps sustain the development of these advanced features. The app is designed to be portable, meaning it can be used across different supported machines as long as the VRAM threshold is met.

What technical details define the local operation of the Portable Computer?

Several key components of the agent system run locally. These include the agent harness, the planner, the tool router, the scheduler, the durable task queue, the local search index, and the sandbox. Each of these elements contributes to the autonomous operation of the AI agent without external dependencies. The local search index allows the agent to access and process data stored on the device. The sandbox provides a secure environment for executing tasks. This architecture ensures that the entire workflow stays on the hardware, enhancing both speed and security. The optimization for RTX GPUs means that the models are tuned to make the best use of the available video memory and processing cores.

1. The agent harness manages the overall operation of the local AI agent.
2. The planner breaks down complex tasks into manageable steps locally.
3. The tool router directs the agent to appropriate local tools and resources.
4. The scheduler organizes the timing and priority of tasks on the device.
5. The durable task queue maintains task persistence even if the system is interrupted.
6. The local search index enables quick retrieval of on-device information.
7. The sandbox creates an isolated environment for safe task execution.

Supported Hardware Configurations for Portable Computer

Hardware Platform	Support Status	Minimum VRAM	Operating System
GeForce RTX GPUs	Available now	24 GB	Windows 10/11
RTX PRO Workstations	Available now	24 GB	Windows 10/11
DGX Spark	Available now	Specified by system	Linux and Windows
DGX Station	Expected soon	N/A	To be announced
Linux RTX PCs	Prior support	24 GB	Linux

These local components work together to provide a seamless experience for users engaging in agentic workflows. The system is built to be efficient on the specified hardware configurations. With the addition of Windows support, more users can benefit from these technical advancements. The post-training of the Qwen 3.8 27B model specifically for this use case demonstrates the effort put into optimizing performance on RTX GPUs. Users can expect reliable operation as long as the VRAM requirement is satisfied.

What implications does this have for the AI agent market and stakeholders?

The availability of local AI agents on consumer and professional Windows PCs could shift the market towards more decentralized AI solutions. Companies like Perplexity and NVIDIA are positioning themselves as leaders in this space by providing tools that reduce dependency on large cloud providers. For stakeholders such as software developers and businesses, this means new opportunities to build applications that respect user privacy by default. Enterprise users with Max subscriptions can deploy these agents in environments where data cannot leave the premises. The cost savings from not using cloud credits for every task could make AI agents more accessible to a wider range of users. This development may encourage other companies to develop similar local-first solutions.

High-end PC owners who have invested in powerful RTX GPUs now have a practical use case for their hardware beyond gaming and traditional computing tasks. This could increase the value proposition of such systems. The partnership between Perplexity and NVIDIA highlights the importance of hardware software co-optimization in the AI field. As more tasks move to local execution, the demand for high VRAM GPUs may rise, benefiting NVIDIA's hardware sales. Overall, the market sees a move towards hybrid systems where local and cloud resources are used based on user preference and task requirements.

What have experts and the company said about the privacy and cost benefits?

Perplexity Portable Computer is a local-first agent. Sensitive data therefore never leaves the device without permission, and local models carry no inference fee: the system is private and cost-effective by construction.Perplexity

This statement underscores the core advantages of the local approach. By keeping data on device, users can trust that their information is protected. The absence of inference fees for local models makes it economically attractive for intensive use. The company announcement highlights these points as fundamental to the design of the Portable Computer. Such features are likely to appeal to users who are wary of cloud based AI services due to privacy policies or recurring costs. The explicit permission for cloud escalation further reinforces user control.

What developments are expected in the near future for this technology?

Support for the DGX Station is anticipated soon, which will further expand the ecosystem of supported hardware. The upcoming addition of Nemotron 3.5 Lightning to the local models will provide users with more options for running the agent. As the technology matures, additional features and optimizations are likely to be introduced. The extension from Linux and DGX systems to Windows PCs indicates a commitment to broadening accessibility. Users can look forward to a more robust local AI agent experience as these updates roll out. The focus remains on maintaining the local-first principle while enhancing capabilities.

The overall trajectory suggests that local AI agents will become more common on personal computers equipped with sufficient hardware. This could lead to a new category of AI powered applications that run natively on Windows. Perplexity's collaboration with NVIDIA sets a precedent for how software and hardware can be integrated for agentic AI. As more users adopt this technology, feedback will likely drive further improvements in performance and usability. The current release is just the beginning of what local agentic AI can offer on high end RTX PCs.

How does the subscription model support the local AI agent feature?

The Pro and Max subscriptions provide access to the Portable Computer, ensuring that the service is sustainable. Enterprise users can benefit from the Max plan for larger scale deployments. This model allows Perplexity to continue developing and supporting the local agent technology. The subscription also unlocks other features in the Perplexity ecosystem, making it a comprehensive offering for AI users. By tying the local agent to subscriptions, the company maintains a revenue stream while delivering value through privacy and cost savings. Users who frequently use agentic tasks will find the subscription worthwhile given the local execution benefits.

The requirement for 24GB VRAM or more ensures compatibility with high performance hardware, which is necessary for smooth operation of the large language models involved. This hardware threshold also positions the product for users who already have invested in top tier NVIDIA GPUs. The Windows app integration makes it easy for existing Perplexity users to enable the feature with minimal setup. Overall, the subscription and hardware requirements are designed to deliver a high quality experience for those who meet the criteria.

In conclusion, the launch expands the reach of local AI agents significantly. It combines the power of NVIDIA RTX hardware with Perplexity's agent technology to offer a private and efficient alternative to cloud based solutions. As support for additional hardware comes online, more users will be able to take advantage of this innovative approach to agentic AI. The focus on local execution without compromising on functionality sets a new standard in the industry.

The detailed integration of local components like the planner and scheduler ensures that the agent can operate independently for extended periods. This independence is crucial for tasks that require continuous processing without interruptions from network issues. The local search index adds value by allowing the agent to utilize personal data securely. Such features make the Portable Computer a versatile tool for a variety of professional and personal applications on Windows PCs.

With the PPLX 27B and other models optimized for this environment, performance is tailored to the local hardware. The benchmark results indicate strong capabilities in local knowledge work tasks. This optimization is key to making local agents practical for everyday use. The collaboration with NVIDIA ensures that the hardware is fully utilized for these AI workloads.