Perplexity partners with Nvidia to launch Portable Computer, a fully

What happened

Perplexity is partnering with Nvidia to launch Portable Computer, a version of its agentic "Computer" platform that runs entirely on hardware users already own, starting with Nvidia's DGX Station desktop. The product eliminates cloud inference entirely, meaning every task the agent performs — browsing, reasoning, executing multi-step workflows — happens on the local machine with no tokens consumed and no per-query billing.

This piece breaks down what Portable Computer is, how the local execution model differs from cloud-hosted alternatives, and what developers and enterprise buyers who are already evaluating agentic AI infrastructure need to understand before making decisions. If you are building on Perplexity's existing "Computer" product or looking at Nvidia's DGX Station as an inference platform, the sections below cover the announcement, what is technically different, who stands to benefit most, and what to monitor as the product matures.

Perplexity and Nvidia announced Portable Computer today, positioning it as a direct extension of Perplexity's existing "Computer" agentic platform. The collaboration pairs Perplexity's agentic software layer — capable of taking multi-step actions across the web and desktop applications — with Nvidia's DGX Station, a high-performance desktop workstation designed for local AI inference. The product's headline promise is zero token costs, because all compute happens on hardware the user already controls rather than routing through a hosted API. Neither company has disclosed pricing for the software license or whether Portable Computer will be bundled with DGX Station purchases.

How it works

The partnership signals that both companies see a market segment that neither fully addressed alone: users who want the capability of a cloud-hosted AI agent but need or prefer to keep data and inference on-premises. Perplexity's "Computer" brand has been building toward agentic autonomy — tasks that span multiple applications and require persistent context — and the DGX Station gives that software a capable local substrate without requiring a data-center footprint.

The core architectural shift is where inference happens. In Perplexity's cloud-hosted "Computer" offering, model calls route to Perplexity's infrastructure and are billed by usage. In Portable Computer, the model runs on the DGX Station's local GPUs, so there is no outbound API call for inference and no per-token accounting. The agent's ability to browse, interact with desktop interfaces, and chain reasoning steps appears to be preserved, but the execution environment is the local machine rather than a remote cluster.

Why it matters

What is not yet fully detailed is which model or models power Portable Computer locally, how updates to those models are distributed, and whether the agentic capabilities are feature-equivalent to the cloud version or a subset. Builders evaluating the product should verify those specifics directly, since the gap between a cloud agent with frequent model updates and a locally deployed agent on a fixed model version is operationally significant for teams that depend on capability improvements over time.

The clearest beneficiaries are organizations with strict data-residency requirements — healthcare, legal, government, and financial services — where sending prompts and context through a third-party cloud API creates compliance exposure. For those teams, a fully local agent that can execute complex workflows without any data leaving the building is a meaningful change in what agentic AI becomes usable for. Researchers and engineers already running Nvidia DGX Station hardware have an obvious low-friction path to try the product without adding infrastructure.

Individual developers and small teams are less obviously served right now. The DGX Station is high-end workstation hardware, so the on-ramp assumes either existing access to that class of machine or willingness to acquire it. The zero-token-cost framing is compelling in theory, but the total cost of ownership calculation depends heavily on whether buyers already have compatible Nvidia hardware sitting underutilized or would be purchasing it specifically for this use case.

Who is affected

Perplexity says Portable Computer is launching today, and the partnership with Nvidia centers on the DGX Station as the initial target platform. Beyond that, specific availability details — whether access is gated to a waitlist, open to all Perplexity users, or restricted to enterprise contracts — have not been disclosed in the announcement. Interested users should check Perplexity's website and the DGX Station product page directly for sign-up or access instructions.

For anyone already operating a DGX Station, the practical starting point is confirming driver and software compatibility requirements, since local model deployment typically depends on specific CUDA versions and driver stacks. Builders should also test whether Portable Computer's agentic capabilities — particularly any web-browsing or desktop-interaction features — require ongoing connectivity to Perplexity's servers for orchestration even if inference itself is local, which would affect air-gapped deployment scenarios.

What to watch next

The most important open question is model governance: how Portable Computer handles model updates, whether users can swap in their own models, and what the support lifecycle looks like for locally deployed versions. Cloud agents benefit from silent, continuous improvements; local deployments require deliberate update processes, and the operational burden of keeping a local agent current is meaningfully different from using a hosted API.

A second thing to track is whether Portable Computer expands beyond Nvidia's DGX Station to other hardware configurations. Limiting initial availability to a specific high-end workstation makes the product easier to support and certify, but it also caps the addressable market significantly. If Perplexity and Nvidia open compatibility to a broader range of Nvidia GPU configurations, that would signal whether Portable Computer is a platform play or a tightly scoped co-marketing effort.