FULL STORY

Perplexity Portable Computer Rolls Out on Linux, Then Windows

Perplexity's local agent product Portable Computer first landed on Linux and DGX Spark, then officially launched on Windows with NVIDIA, requiring RTX GPUs with at least 24GB VRAM for fully local agents.

2026-09-04 ~ 2026-09-15 · 2 episodes · 17 posts

Episode 1 · Perplexity Portable Computer Lands on Linux and DGX Spark (2026-09-04, 3 posts)

Perplexity's fully local agent stack, Portable Computer, is now available on Linux with support for NVIDIA RTX GPUs with 24GB+ VRAM, and has also landed on NVIDIA's DGX Spark.

Episode 2 · Perplexity Brings Portable Computer to Windows RTX PCs with NVIDIA, HP and Autodesk (2026-09-14, 14 posts)

On September 14, Perplexity officially launched Portable Computer on Windows in partnership with NVIDIA: the harness, agents and models all run locally on the user's machine, handling local files and connected apps without sending tasks to the cloud, keeping sensitive data on-device while allowing cloud resources when needed. Perplexity then announced a preinstall partnership with HP and an Autodesk integration, pushing the local agent further into the PC ecosystem.

Confirmed

  • Portable Computer is available in the Windows app, first supporting Windows and Linux; it previously supported DGX Spark and other platforms, and Perplexity released the Personal Computer agent in February.
  • Hardware requirement is an NVIDIA RTX GPU with 24GB or more VRAM, covering GeForce RTX cards and RTX PRO workstations.
  • Perplexity states Portable Computer is the local-first version of Computer: tasks can run fully on-device with local models, privacy data stays local and does not consume cloud credits.
  • According to @testingcatalog, the update adds support for local MCP and scheduled tasks.
  • CEO Arav Srinivas announced an expanded NVIDIA partnership, confirmed by NVIDIA's official accounts.
  • Perplexity will preinstall Computer on the HP ZBook Ultra G3a, expanding to more HP devices later; users can run multi-step tasks via a unified interface, with agents connecting hundreds of tools.
  • With Autodesk, the Computer agent connects to Revit via Autodesk's MCP server, reading open Revit building model data for reports and view exports.
  • A Microsoft executive shared the news; @jeffgrimes9 relayed that Windows is becoming the platform for local AI, helping users work while keeping data control and lowering costs.

Why it matters

  • Local execution keeps sensitive files and app data off the cloud, addressing enterprise and consumer privacy and compliance concerns.
  • Not consuming cloud credits ties localization directly to cost control—heavy agent users save on cloud API spend, a likely selling point of the local-first strategy.
  • The 24GB+ VRAM requirement sets a high bar, reflecting dependence on high-end consumer GPUs and potentially driving demand for RTX cards, workstations and preinstalled HP PCs.
  • Local MCP and scheduled tasks enable agents to automate local tools on a schedule; Revit integration extends local agents into architecture and other vertical workflows.
  • Involvement from NVIDIA, Microsoft, HP and Autodesk signals that local agent execution is becoming the next competitive battleground for PC platforms.