Local inference is becoming a product category, not just a benchmark table.

Perplexity launched Portable Computer on Windows RTX PCs today. The orchestrator LLM, subagent LLM, and the full agent harness all run on-device. No cloud dependency for most tasks. Zero token costs when running locally.

The hardware requirement: an NVIDIA RTX GPU with at least 24GB of VRAM. RTX 3090 and newer. The default local model is Qwen 3.8 27B, post-trained specifically for Perplexity Computer’s harness. Frontier cloud models are called on-demand when reasoning demands them.

This is not another chat wrapper running locally. It is a full agentic stack with task planning, subagent delegation, file access, and connected apps running on the hardware you own.

The competitive landscape: Ollama does local model serving. LM Studio and http://Jan.ai add desktop UIs. AnythingLLM adds RAG. None of them ship a production-grade agent harness with orchestrator and subagent tiers as a first-class primitive.

NVIDIA has over 100 million RTX GPUs in market. The share with 24GB or more is a fraction today, but each new RTX generation pushes 24GB to mainstream price points.

My read: the bottleneck shifts from inference cost to harness quality. Perplexity is positioning as the harness, not the model. That is a real strategic bet.

Jan - Open-Source ChatGPT Replacement

Source: https://www.jan.ai/

Meet Jan

Personal Intelligence that answers only to you

An open model ecosystem

Jan Agent

The core agent, distributed separately — run it on your own VM or container.

Tokamak

Router, fusion model, and governance/audit — the self-hosted backend Jan agents connect to.

Over 4 million downloads

Jan is built in public

We believe AI should be open, and grow through the people who build and use it

All the tools you need to make Jan yours

1

Models

Choose from open models or plug in your favorite online models.

ChatGPTOpenAI

ClaudeAnthropic

GeminiGoogle

LlamaMeta

MistralMistral AI

QwenAlibaba

DeepSeekDeepSeek

GemmaGoogle

KimiMoonshot AI

2

MemoryComing Soon

Your context carries over, so you don’t repeat yourself. Jan remembers your context and preferences.

Things Jan keeps in mind

• •Minimalist UI tasted
• •Currently on a portfolio refresh
• •Wants brief, to-the-point answers
• •Frequent Figma/prototyping questions
• •Dark-mode sharer
• •Curious about type trends (Mostly harmless)

Ask Jan anything

+6.6Mdownloads, Free & Open source

Perplexity (@perplexity_ai): Portable Computer is now available on Windows PCs with @NVIDIA RTX GPUs.

Run the harness, agents, and models locally on your PC.

Work with local files and connected apps without sending tasks to the cloud. Use frontier cloud models when needed.