How to Install and Use Perplexity Portable Computer on NVIDIA DGX Spark

01

Step 1: confirm the launch requirements

At launch, Portable Computer is for eligible Perplexity Pro or Max users on NVIDIA DGX Spark running Linux. For the product-level context, see the Perplexity Portable Computer overview.

02

Step 2: install through the Perplexity app

Perplexity says DGX Spark users can set up Portable Computer through the Perplexity app with a one-click installation flow. Use the current product-provided installer rather than copying unofficial shell commands.

The launch material does not publish a universal daemon name, port layout or manual package sequence, so those details should not be invented in a setup guide.

03

Step 3: choose the local model

The launch model choices are Qwen 3.8 27B and PPLX 27B. Perplexity positions PPLX 27B as a post-trained local-work model that can complete tasks on-device and decide when an authorized cloud capability may be useful.

04

Step 4: begin with a local-only task

Before connecting external services, give Portable Computer a task that can be completed from local files—for example summarizing a folder, searching a private document set or classifying local material. Confirm that the task can complete without a remote step.

Perplexity says local work does not consume credits, so this is also a useful baseline for understanding which workloads remain on-device.

05

Step 5: test cloud escalation deliberately

Next, ask for something that genuinely requires current web information or a remote model. Portable Computer can escalate selected steps to Perplexity search, research, browser tools or frontier models, and Perplexity says the user is asked for permission before device content is sent to the cloud.

Read that authorization step carefully when the local task contains confidential documents, private code or sensitive business material.

06

Step 6: add connectors only when needed

Perplexity lists Google Drive, Gmail, Slack and GitHub among supported integrations. Connecting one creates an intentional boundary between local processing and an external service, so start with the minimum access needed for the workflow.

A workflow that reads or writes through Gmail, Drive, Slack or GitHub should not be described as fully on-device simply because its orchestration began locally.

07

Step 7: validate the hybrid workflow

1. Process one local document without cloud escalation
2. Request current web information and confirm the authorization step
3. Connect one external app and test a limited read action
4. Identify which steps stayed local and which crossed to a remote service
5. Repeat with the other launch local model if you want to compare behavior

This tests Portable Computer’s defining local/cloud boundary rather than only confirming that the application launches.

08

Step 8: keep sandboxing and human review in the loop

Perplexity says code and tool execution uses isolated sandboxes with controlled file and app access. Continue reviewing destructive file operations, repository changes, messages and external writes, and use backups or version control where appropriate.

09

Do not assume future Windows and RTX behavior

Windows plus broader GeForce RTX and RTX PRO support are announced as coming later. Until those builds ship, third-party claims about exact supported cards, VRAM minimums, installer commands or rollout dates should not be treated as official Perplexity requirements.

10

Bottom line

The launch setup is intentionally product-driven: eligible account, DGX Spark Linux, current Perplexity app, local-model choice, a local-only validation task, then permissioned cloud capabilities and connectors. Add each external boundary only after the local baseline works.