Particle.news

Perplexity Releases Portable Computer to Run Its Agent Locally

It aims to keep data and model inference on users' NVIDIA‑equipped machines by asking permission before any cloud call, so cloud inference costs only apply when users approve.

Overview

  • Perplexity announced Portable Computer on Tuesday, Aug. 25, 2026, and made it available immediately to paid Pro, Max, Enterprise Pro and Enterprise Max subscribers.
  • The product runs the agent on the user’s machine first and asks for explicit approval before sending any individual step to a cloud model, and Perplexity says work fully completed on‑device does not consume cloud billing credits.
  • Portable Computer ships with a local inference engine, a deterministic orchestrator/harness, an OS‑level sandbox for safe tool use, and connectors for Google Drive, Gmail, Slack and GitHub so agents can work with local files and services.
  • At launch it supports Qwen 3.8 27B and Perplexity’s post‑trained PPLX 27B with NVIDIA’s Nemotron 3.5 Lightning expected later, and users may bring their own models or inference servers.
  • The product requires NVIDIA DGX Spark or PCs with NVIDIA RTX GPUs with at least 24GB of VRAM, which makes adoption costly and targets developers, prosumers and organizations; an unconfirmed report that NVIDIA may invest in Perplexity could deepen that hardware–software tie.