Skip to content
All tool news

Tool desk · Updated daily

Tool news·Perplexity·

Perplexity Launches Portable Computer on NVIDIA DGX Spark for Local AI.

Perplexity Portable Computer runs its agent stack locally on NVIDIA DGX Spark, with approval required before using cloud models.

CW

Create With tool desk

1 source checked · 3 min read · 3 sections

ShareLinkedIn

Perplexity is moving its Computer agent off the cloud, at least for customers with the right hardware. The company has launched Portable Computer on NVIDIA DGX Spark, a local deployment that runs its orchestration model, subagent model and agent harness directly on the user’s machine.

The release targets a persistent concern around AI agents: sending sensitive files and context to remote infrastructure. According to Perplexity’s launch announcement on LinkedIn, Portable Computer has no cloud dependency for its standard runtime. Documents remain on the DGX Spark while the local models plan and execute tasks.

That makes the product notably different from the cloud model integrations Perplexity has recently added to Computer. Those releases have included DeepSeek V4 Pro in U.S.-hosted workflows and Grok 4.6 across Answer and Computer. Portable Computer instead makes local processing the default, while retaining a controlled route to more capable cloud systems.

What Runs Locally

Portable Computer installs and runs a post-trained PPLX 27B model on the customer’s hardware. Perplexity says Qwen 3.8 27B is also available, with support for NVIDIA Nemotron 3.5 Lightning planned.

The local runtime includes three important layers:

  • The orchestrator LLM, which interprets the request and coordinates the task
  • Subagent LLMs, which handle delegated parts of the workflow
  • The agent harness, which manages the broader execution process

Keeping all three layers local matters. A product that runs only its primary language model on-device could still transmit prompts, intermediate steps or documents through a cloud-based orchestration service. Perplexity’s stated architecture is broader, covering the system that plans and operates the agent as well as the models themselves.

The release is limited to text guidance. Perplexity also describes the system as user-gated and PII-flagged, although its announcement does not provide detailed technical documentation for those controls. In practice, the design is intended to help users identify personal information and retain authority over actions that could move a task outside the local environment.

Cloud Escalation Requires Approval

Local models may not be sufficient for every assignment. When Portable Computer determines that a task needs what Perplexity calls “frontier reasoning,” the orchestrator can propose routing it to a frontier model in the cloud.

That handoff is not automatic. The system asks the user for approval first, and Perplexity says the escalation is limited to text guidance. Sensitive documents stay on the device, according to the company. The distinction gives businesses a middle path between fully local AI, which can offer greater data control, and cloud-hosted frontier models, which may perform better on difficult reasoning tasks.

The approval step will be especially relevant for teams working with client records, internal strategy documents, financial material or other information they do not want uploaded by default. Still, Perplexity has not disclosed in the announcement exactly what text is sent during an approved escalation, how PII flags are surfaced or whether administrators can establish organization-wide policies. Buyers assessing the feature for regulated work will need those details before treating it as a compliance solution.

Portable Computer is available now to Perplexity Pro and Max subscribers who have an NVIDIA DGX Spark. No separate price was announced, and the hardware requirement makes this a specialized deployment rather than a local mode for ordinary laptops.

For small businesses and builders, the immediate trade-off is clear. Portable Computer provides more control over where agent workloads run, but requires dedicated NVIDIA hardware and a qualifying subscription. It is most relevant to teams that have enough sensitive, repeatable AI work to justify maintaining local infrastructure.

Frequently asked questions

What is Perplexity Portable Computer on NVIDIA DGX Spark?

It is a fully local version of Perplexity Computer. Its orchestrator LLM, subagent LLMs and agent harness run on an NVIDIA DGX Spark, allowing standard tasks and sensitive documents to remain on the user’s hardware.

Does Portable Computer ever send tasks to the cloud?

It can request cloud assistance when the local orchestrator determines that frontier reasoning is needed. The user must approve that handoff first, and Perplexity says it is PII-flagged, limited to text guidance and designed to keep sensitive documents on-device.

Who can use Portable Computer?

Portable Computer is available to Perplexity Pro and Max subscribers who own an NVIDIA DGX Spark. Perplexity has not announced support for conventional laptops or other local hardware in the provided release information.

Sources

1 checked

How we cover tool news: Create With's tool desk drafts these reports with AI from the sources listed above and checks them against those sources before publishing.

Worth passing on?

ShareLinkedIn

Go deeper on Perplexity

Related reading, watching and going.

Everything on Perplexity →

The briefing

16 Jun 2026

How Claude Code Lets One Person Replace SEO Teams Fast

Discover how Claude Code lets one non-technical user automate SEO — from keyword discovery to publishing — replacing SEO teams easily without writing code.

Read →

The briefing

31 Mar 2026

Vibe Marketing Review: Can AI Agents Really Automate Marketing?

An honest review of vibe marketing with AI agents. Learn if automated marketing actually works for solo founders and small teams. Plus: Claude Code leak, Factory Missions.

Read →

The briefing

3 Mar 2026

Perplexity Computer Platform & Notion AI Agents Review 2025

Perplexity's 19-model AI platform, Claude Code mobile control, iPhone offline AI with Qwen 3.5, and Notion's custom agents are transforming AI automation.

Read →
31 min

Conference

How to Get Your Product Recommended by Google and ChatGPT

Tanya Van Gastel presents the APEX framework for making products discoverable in AI-powered search engines like ChatGPT and Google's AI Overviews. She explains how AI is fundamentally changing product discovery and buyer behavior, then breaks down five critical components: Authority (being cited as a solution), Presence (visibility across search platforms), Expertise (creating high-quality money pages), Endorsements (off-site validation), and Extractability (making content easy for AI to process). The talk emphasizes quality over quantity, with data showing 50% of SaaS SEO revenue comes from fewer than 10 high-performing pages.

Watch now ↗

Latest tool news

What else changed this week.

All tool news
VercelDigest

What Vercel Shipped in Late October

Vercel shipped quicker static deployments, four AI Gateway models, more free domain choices and new CLI controls in late October. The changes cover developers testing sites…

Vercel

Vercel Adds Submit_feedback to Its MCP

submitfeedback is a new Vercel MCP tool that lets agents and harnesses send bug reports, friction reports and feature requests directly to the Vercel team. Feedback can include…

The Create With Briefing

Don't watch forty changelogs. Read one email.

Every Tuesday: the tool changes worth knowing, real business use cases, and what's on near you. Free, unsubscribe any time.