Perplexity's Portable Computer local AI agent comes to Windows RTX PCs

Perplexity's Portable Computer — a local, on-device version of its Perplexity Computer agent — is now available on Windows PCs equipped with NVIDIA RTX GPUs, extending a feature that previously ran only on NVIDIA DGX Spark systems and RTX PCs running Linux.

What's new

Portable Computer plans and carries out multistep tasks directly on a user's PC rather than in the cloud, using a local model (NVIDIA highlights Qwen 3.8 27B as an example) that is post-trained to work with Perplexity Computer and optimized for NVIDIA RTX GPUs. Perplexity is rolling it into the Perplexity app for Windows on compatible NVIDIA GeForce RTX PCs and NVIDIA RTX PRO Workstations.

Key details from NVIDIA's announcement:

• Requires an NVIDIA GeForce RTX or RTX PRO GPU with 24GB or more of VRAM; NVIDIA DGX Station support is "expected to come soon."
• Connectors for Microsoft Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub let the local agent work across a user's existing files and apps.
• For tasks that need heavier reasoning, Portable Computer can identify when a job requires cloud support and asks the user for permission before sending any information off-device.
• Local completions don't consume Perplexity Computer cloud credits, and NVIDIA notes that "sensitive information stays on device, and locally completed work doesn't consume Perplexity Computer credits."

NVIDIA gives concrete use cases: reviewing and triaging open GitHub pull requests, tracing recurring investment fees across years of brokerage statements without a document ever reaching a chatbot, and analyzing a startup's funnel data locally to explain a drop in activation, then posting the findings straight to Slack.

Context

This is an expansion, not a first launch — Perplexity's local-agent push started on NVIDIA DGX Spark hardware and Linux RTX PCs. Bringing it to mainstream Windows RTX gaming and workstation GPUs is the step that puts on-device agentic AI in front of a much larger consumer and prosumer hardware base, rather than specialized AI workstations alone.

Why it matters

The pitch here is privacy and cost: work that touches sensitive local files (tax documents, internal financial data, proprietary code) never has to leave the device or spend cloud credits, while the app still offers a path to heavier cloud reasoning when a task calls for it. For NVIDIA, every one of these integrations turns existing RTX GPU inventory into an argument for on-device AI capability, reinforcing local inference as a distinct product category alongside cloud AI assistants rather than a stopgap for offline use.