K

GUIDE

Getting started with TARX

Install TARX, set up your first Space, and run your first local AI conversation — all in under five minutes.

Last updated · March 20, 2026

guide2 min read·by TARX

TARX runs AI on your hardware. No cloud accounts, no API keys, no data leaving your machine. Here's how to get started.

Download and install

Head to tarx.com/download and grab the macOS DMG. Open it, drag TARX to Applications, and launch.

On first launch, TARX will download the Mind model (~4.7 GB). This is a one-time download. The model runs entirely on your Mac's GPU using Metal — no internet required after the initial setup.

Your first conversation

Open TARX and start typing. That's it. There's no account creation, no sign-in, no onboarding wizard.

TARX uses two inference layers:

  • Soul — Apple's on-device Foundation Model. Handles fast, lightweight reasoning. Always available, always private.
  • Mind — A Qwen3-8B model running locally via llama-server. Handles deeper analysis, code generation, and complex queries.

Both run in parallel. You never pick a mode. TARX automatically synthesizes the best response.

Create a Space

Spaces are how you organize your work in TARX. Each Space has its own conversation history, files, and memory.

Tap the + button in the sidebar, give your Space a name, and you're in. Everything in a Space is stored locally in SQLite on your device.

Some ideas for Spaces:

  • A project you're building
  • Research on a topic
  • A client engagement
  • A personal journal

Add files

Drag files into a Space to give TARX context. TARX indexes them using local embeddings (nomic-embed-text) and chunks them for retrieval. Your files never leave your machine.

Supported: text, code, markdown, PDF, images. TARX will reference these files when you ask questions about them.

Connect to the Supercomputer

For queries that need more power — long documents, complex reasoning, multi-step analysis — TARX can connect to the Supercomputer at compute.tarx.com.

This is opt-in. Go to Settings and toggle Supercomputer access. When enabled, TARX sends the query (not your files) to the Supercomputer for Mind-class inference and streams the result back.

You can disable this at any time. With the privacy flag on, nothing leaves your hardware — ever.

What's next


TARX
«