The 60-second briefing
Perplexity says Portable Computer is now available in its Windows app on PCs with NVIDIA RTX GPUs, with on-device inference requiring at least 24GB of VRAM. The product runs its agent harness and models locally, works with local files and connected apps, and can use frontier cloud models when needed.
What changed
This is a material availability expansion of Portable Computer, which Perplexity first introduced on NVIDIA DGX Spark and Linux. The new Windows path makes its local-first agent workflow available to a broader class of RTX workstations. Perplexity also says the Windows release adds local MCP support and scheduled tasks, so users can connect their own tools and run recurring work on the PC.
Why it matters
The important shift is where the work happens. Perplexity describes a workflow in which local inference and the harness can keep private files and routine agent work on the machine instead of sending every step to the cloud. If a task needs web access or stronger reasoning, the user can still use cloud models - but the product says that escalation is optional and user-controlled.
Availability
Verified: the official Perplexity account announced the Windows RTX availability at 2026-09-14 15:03:31 UTC. The companion setup post says on-device inference requires an NVIDIA RTX GPU with 24GB of VRAM or higher. Availability is public but hardware-constrained; the official post does not state a new subscription price in this announcement.

