Core Announcement: Portable Computer Local Agent Now Available on Windows RTX PCs
Perplexity’s local agent platform, Portable Computer, is now officially open to Windows RTX PC users as of September 14, 2026.
Key Technical Specifications and Availability:
- Release Date: September 14, 2026
- Supported Hardware: PCs with NVIDIA GeForce RTX graphics cards and NVIDIA RTX PRO workstations
- Minimum VRAM: 24GB or higher (for both GeForce RTX and RTX PRO)
- Built-in Model: Qwen 3.8 27B (fine-tuned for Perplexity Computer and optimized for NVIDIA RTX GPUs)
- Deployment: No complex local AI setup required—ready out of the box
- Cloud Fallback: Automatically detects high-complexity tasks and requests user authorization before sending data to the cloud
Previously available on NVIDIA DGX Spark systems and Linux-based RTX PCs, this marks the first large-scale expansion of the feature to Windows users.
Core Capabilities and Technical Architecture
Portable Computer is the on-premises version of Perplexity Computer, designed to keep sensitive data on the local machine while executing multi-step tasks. Powered by NVIDIA GPU acceleration, it uses local models to analyze data, consolidate information across multiple documents, and automate repetitive workflows.
A notable technical surprise: the built-in Qwen 3.8 27B model has 27 billion parameters, significantly larger than many consumer-grade local models (typically 7B-13B). Previously, running 27B-scale models in real-time on consumer RTX hardware was uncommon outside高端 workstations. The fact that Windows RTX PCs can now handle this model size reflects substantial progress in GPU architecture and model inference optimization.
Authorization Workflow: When a task requires higher-level reasoning, the system automatically identifies the need for cloud resources and prompts the user for explicit consent before any data leaves the device—giving users full control over data flow.
Enterprise Workload Integration
Portable Computer includes connectors for major business applications:
- Microsoft: Outlook, OneDrive, Word
- Google: Google Drive, Gmail
- Collaboration & Dev: Slack, GitHub
These integrations enable the agent to read and analyze data across platforms while keeping all processing local—critical for users who cannot or prefer not to upload sensitive files.
Example Use Cases (Facts from Source Material)
| Domain | Task Description | Data Flow |
|---|---|---|
| Engineering | Read GitHub pull requests; categorize by status; identify outdated documents and propose edits | Fully local; code never leaves device |
| Finance | Process brokerage statements, consolidated 1099 tax forms, and tax returns; identify assets with high fees/tax drag with document annotations | Fully local; files remain on machine |
| Startups | Analyze local conversion funnel data; identify user drop-off points; send findings directly to Slack channels | Fully local; only conclusions are pushed |
Target Users and Recommendation
Users Ready to Try Immediately:
- Owners of GeForce RTX 4090, 4080 Super, or higher-tier cards
- Financial professionals processing sensitive documents (requires 24GB+ VRAM)
- Engineering teams collaborating on GitHub who need PR review assistance
- Users with strict data privacy requirements who avoid cloud-based document processing
Users Should Wait:
- RTX PC owners with less than 24GB VRAM (e.g., RTX 4070 12GB variants)
- Users not yet updated to the latest Windows client version
- General users needing only lightweight local inference—lower门槛 tools may suffice
NVIDIA DGX Station workstation support is “coming soon”—DGX Spark system users already have access to the complete workflow.
Final Thoughts
Portable Computer represents a shift from prototype to practical tool: when a 27B-parameter model can reliably run on ordinary RTX PCs and handle real-world workflows, local AI’s operational value truly emerges. Future competition will increasingly hinge not just on model scale, but on how deeply models are coupled with local hardware ecosystems.