What does Perplexity Portable Computer cost?

Portable Computer is included with an active Perplexity Pro, Max, Enterprise Pro or Enterprise Max subscription. Work that finishes on your own machine consumes no billing credits, because local inference has no per-token API fee. You supply the hardware, so the running cost is your own electricity and the NVIDIA GPU you already own.

What hardware do you need to run Portable Computer?

Portable Computer needs an NVIDIA GPU with at least 24GB of VRAM, which is roughly a GeForce RTX 3090 or newer. It runs on the desktop-sized NVIDIA DGX Spark workstation, built on the Grace Blackwell GB10 platform with 128GB of unified memory, and on other Linux machines running DGX OS or Ubuntu on ARM or x64.

Does Portable Computer send my data to the cloud?

Portable Computer starts every task on the device and never escalates silently. When a step needs a stronger model, it runs a PII classifier over the outgoing context and shows exactly what would leave the machine before asking permission. The remote model returns guidance only — it never gets access to your local files or tools.

Can I run Portable Computer on Windows?

Not yet. Portable Computer launched on Linux only, and Perplexity says Windows support arrives in September 2026. Until then the supported setups are NVIDIA DGX Spark with DGX OS, or an Ubuntu machine on ARM or x64 with a qualifying NVIDIA GPU.

PPLX 27B is Perplexity's own post-trained build of Qwen3.8-27B, published on Hugging Face. The checkpoint went through RFT and SDPO post-training, was quantized to NVFP4, and ships with draft tensors baked in for DFlash2 speculative decoding, so it serves faster on DGX Spark and GB10 hardware.

Perplexity · 2026-08-25 · major

Portable Computer — Perplexity's agent runs fully on your own GPU

Perplexity Portable Computer runs its whole agent stack — orchestrator, subagents and harness — on hardware you own. Local work uses no billing credits, and the agent asks permission before sending any step to a cloud model.

Illustration for The Register's report on Perplexity's local-first Portable Computer agent — The Register

Perplexity's full agent runtime now runs on a DGX Spark or a 24GB RTX GPU, with no cloud dependency.

Key specs

Local knowledge work bench	82.6%
Browse comp	66.7%

Quick facts

Maker	Perplexity, with NVIDIA
Local models	Qwen3.8-27B or PPLX 27B
Hardware	NVIDIA DGX Spark, or an RTX GPU with 24GB+ VRAM
Operating system	Linux at launch; Windows in September 2026
Plans	Pro, Max, Enterprise Pro, Enterprise Max
Billing	Local work uses no credits
Coming next	NVIDIA Nemotron 3.5 Lightning

What is it?

Portable Computer moves the entire Perplexity Computer runtime — the orchestrator LLM, the subagent LLM and the agent harness — onto hardware you own. Perplexity launched it with NVIDIA on 25 August 2026 for Pro, Max, Enterprise Pro and Enterprise Max subscribers. The first build is Linux-only and needs an NVIDIA GPU with 24GB of VRAM or more.

How does it work?

Every task starts on the device, running either Qwen3.8-27B or PPLX 27B, Perplexity's post-trained NVFP4 build of that model with DFlash2 speculative decoding. When a step needs a stronger model, a PII classifier scans the outgoing context, the app shows exactly what would leave the machine, and nothing is sent until you agree. The cloud model replies with guidance only and never touches local files or tools.

Why does it matter?

Running the agent locally removes the per-token API bill and keeps private context off remote clusters, which is the blocker for teams that cannot ship documents to a cloud endpoint. Perplexity reports 82.6% accuracy on its 53-task Local Knowledge Work Bench and 66.7% on BrowseComp with the 27B model, so the local setup is not a toy fallback.

Who is it for?

privacy-sensitive teams and DGX Spark or RTX owners

Frequently asked questions

What does Perplexity Portable Computer cost?: Portable Computer is included with an active Perplexity Pro, Max, Enterprise Pro or Enterprise Max subscription. Work that finishes on your own machine consumes no billing credits, because local inference has no per-token API fee. You supply the hardware, so the running cost is your own electricity and the NVIDIA GPU you already own.
What hardware do you need to run Portable Computer?: Portable Computer needs an NVIDIA GPU with at least 24GB of VRAM, which is roughly a GeForce RTX 3090 or newer. It runs on the desktop-sized NVIDIA DGX Spark workstation, built on the Grace Blackwell GB10 platform with 128GB of unified memory, and on other Linux machines running DGX OS or Ubuntu on ARM or x64.
Does Portable Computer send my data to the cloud?: Portable Computer starts every task on the device and never escalates silently. When a step needs a stronger model, it runs a PII classifier over the outgoing context and shows exactly what would leave the machine before asking permission. The remote model returns guidance only — it never gets access to your local files or tools.
Can I run Portable Computer on Windows?: Not yet. Portable Computer launched on Linux only, and Perplexity says Windows support arrives in September 2026. Until then the supported setups are NVIDIA DGX Spark with DGX OS, or an Ubuntu machine on ARM or x64 with a qualifying NVIDIA GPU.
What is PPLX 27B?: PPLX 27B is Perplexity's own post-trained build of Qwen3.8-27B, published on Hugging Face. The checkpoint went through RFT and SDPO post-training, was quantized to NVFP4, and ships with draft tensors baked in for DFlash2 speculative decoding, so it serves faster on DGX Spark and GB10 hardware.

Try it

https://huggingface.co/perplexity-ai/pplx-computer-qwen-3-8-27b-dflash2-20260824