Perplexity launches Portable Computer for local-first AI

Perplexity has launched Portable Computer, a version of its Computer agent that runs its core stack on a user's own machine. The product is available now to Pro and Max subscribers on NVIDIA DGX Spark systems running Linux. Support for NVIDIA RTX PCs and Windows is planned but not yet available.

The local stack includes the model, orchestrator, planner, tool router, scheduler, durable task queue and search index. Users can choose Qwen 3.8 27B or PPLX 27B, Perplexity's post-trained version of that model. NVIDIA's Nemotron 3.5 Lightning is due to join the model picker later. The DGX Spark hardware combines a 20-core Arm CPU, an NVIDIA GPU and 128 GB of unified memory.

Perplexity says files, conversations and task traces remain on the device by default. Local work does not consume Perplexity credits. When a task needs current web information, browser use, connected apps or stronger reasoning, the agent can request permission to use Perplexity's cloud services or more than 15 frontier models. The company says content is not sent from the device until the user approves that step. Local dictation is also supported, so audio can be transcribed without being uploaded.

Portable Computer can work with Google Drive, Gmail, Slack and GitHub through connectors. Perplexity gives the example of comparing GitHub issues with emailed bug reports locally, then sending a Slack summary. Code and tool execution take place in isolated sandboxes with controlled access to files and connected applications.

The launch matters because it gives knowledge workers another way to balance privacy, capability and cost. Sensitive code or documents can stay local while selected parts of a workflow reach cloud models when necessary. That is more flexible than choosing between an entirely local assistant and a fully cloud-hosted agent.

There are important limits. The first release requires costly specialist hardware, and Perplexity's claims about handling most tasks locally and benchmark performance come from its own research rather than independent testing. Even so, the product is a concrete sign that smaller open models and desktop AI systems are becoming capable enough to run persistent, tool-using workflows outside a data center.