Perplexity Brings Local AI Agent to Windows, Requires RTX GPUs with 24GB VRAM

Perplexity Portable Computer Expands to Windows RTX PCs

Perplexity has released Portable Computer for Windows through its existing Windows app, partnering with NVIDIA to bring the local AI agent platform beyond Linux-based systems1

1

. The move addresses a key accessibility gap, though strict hardware requirements remain in place. Users must own an NVIDIA RTX GPU with at least 24GB of VRAM, alongside a Perplexity Pro or Max subscription1

1

. Portable Computer was initially launched on August 25 for Linux on DGX Spark systems, which carried a $4,699 price tag1

1

2

2

. This Windows release eliminates the need for separate DGX Spark hardware, though upgrading with a used qualifying card may still cost less than the Founders Edition setup1

1

.

How Portable Computer Differs from Cloud-Based AI Agents

Portable Computer represents a local version of Perplexity Computer, the company's agent designed for multistep AI tasks1

1

. The agent can plan, execute subtasks through connectors and tools, and deliver results beyond simple chat responses1

1

. Unlike cloud-based alternatives from OpenAI and Anthropic, this on-device AI agent runs entirely on the user's PC with GPU acceleration from NVIDIA hardware3

3

4

4

. Local execution delivers two critical advantages: users can hand off tasks without privacy concerns since sensitive data never leaves the device, and on-device work does not consume Perplexity Computer credits, making operations more cost-effective2

2

. When advanced workflows demand more powerful reasoning, Portable Computer can escalate tasks to cloud-based models, but only with explicit user permission1

1

3

3

.

24GB VRAM Requirement Limits Eligible Hardware

The 24GB VRAM requirement creates a significant hardware barrier that cuts across NVIDIA's consumer and professional lineups. Qualifying GeForce RTX cards include the RTX 3090 and 3090 Ti with 24GB, the RTX 4090 with 24GB, and the RTX 5090 with 32GB1

1

. Professional RTX PRO Workstations that meet the threshold span the RTX PRO 4000 at 24GB, RTX PRO 4500 at 32GB, RTX PRO 5000 at 48GB or 72GB, and RTX PRO 6000 at 96GB1

1

. None of these GPUs qualify as affordable for mainstream users, though they remain considerably cheaper than a DGX Spark setup2

2

. Aravind Srinivas, CEO of Perplexity, described the release as bringing "unmetered local intelligence on every Windows PC running on NVIDIA hardware and Perplexity harness"1

1

. NVIDIA DGX Station support is expected soon3

3

.

New Features and Model Support

The Windows release introduces scheduled recurring tasks and local MCP servers for desktop apps1

1

. Connectors extend functionality across Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack, and GitHub, integrating the agent into existing workflows3

3

. The app simplifies setup with a dropdown menu for downloading local models with one click1

1

. Qwen 3.8 27B serves as the primary local model, optimized for NVIDIA RTX GPUs1

1

3

3

. However, the post-trained PPLX 27B version remains exclusive to DGX Spark systems, while Linux and Windows machines with RTX GPUs can only run AI models locally using the standard Qwen 3.82

2

. Perplexity is working to add support for NVIDIA Nemotron 3.5 Lightning on both platforms2

2

.

Practical Applications Across Industries

Portable Computer enables concrete use cases that keep sensitive data on-device while automating complex workflows. Engineers can review open GitHub pull requests, organize them by status, identify next steps, flag outdated documentation, and submit proposed updates for review3

3

. Finance professionals can point the agent at two years of brokerage summaries, consolidated 1099s, and tax returns to trace recurring holdings creating avoidable fees and tax drag, with every figure cited to exact files and pages without documents reaching external servers3

3

. Startups can ask why activation metrics went flat, with the agent analyzing funnel exports locally to find where new signups drop off between install and first completed task, then posting insights directly to Slack channels3

3

. These multistep tasks demonstrate how AI-capable PCs can handle work that traditionally required manual coordination across multiple tools and files.

What This Signals for Local AI Development

NVIDIA announced in a September 3 post that RTX Spark Windows PCs from Lenovo and Acer are expected in October, with two additional local agents—Hermes Agent and OpenClaw—receiving the same simplified setup1

1

. This timing suggests Portable Computer serves as more than a standalone release. The convergence of hardware availability and software support indicates NVIDIA's broader strategy to position RTX machines not just as systems that run AI models locally, but as computers capable of autonomous work4

4

. Watch for whether the 24GB VRAM requirement drops as models become more efficient, potentially opening access to mid-range RTX GPUs. Monitor how enterprises respond to local execution capabilities, particularly in regulated industries where cloud credits and data transmission create compliance challenges. The shift from $4,699 dedicated hardware to existing RTX PCs marks a critical accessibility threshold, even if the GPU requirements remain steep for consumer adoption.