Perplexity's local agent lands on Windows RTX PCs, tokens-free
Run Qwen 3.8 27B on a 24GB RTX card: files, email and GitHub stay on-device and local runs cost zero credits. Plus OpenClaw 2.0 goes multi-agent.
Copy markdown
The unlock: your RTX card, not a $4,699 DGX Spark
When Portable Computer launched Aug 25 it needed NVIDIA's $4,699 DGX Spark; now it runs on any GeForce RTX or RTX PRO card with 24GB+ of VRAM. If you already own a 4090-class GPU, the local agent is a download away, not a hardware purchase.
Local work costs zero tokens and zero credits
Everything the agent does on-device — analyzing files, joining data across documents, running recurring chores — sends nothing to the cloud and burns no Computer credits. Only cloud escalation dips your balance (Max ships 10,000 credits/mo; Pro carries none), so heavy local runs cost you electricity, not API spend.
It plugs into the tools you already work in
The agent connects to Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub, so it can triage your real inbox, files and repos locally. It's a private assistant that can read your codebase and mail without any of it leaving the machine.
The catch: Windows, RTX, 24GB — and Qwen underneath
You need Windows on a GeForce RTX or RTX PRO GPU with at least 24GB of VRAM; laptops and smaller cards are out. The local brain is Qwen 3.8 27B post-trained by Perplexity, not a house model, and demanding reasoning still escalates to the cloud.
Elsewhere: OpenClaw 2.0 turns one agent into a team
The open-source agent that kicked off the autonomous-agent craze shipped 2.0 with multi-agent collaboration, upgraded memory, simpler setup and a redesigned browser UI. Self-hosters get parallel agents coordinating on a task without wiring it up by hand.