Getting started with LU Labs

Download & install

Grab the LU Bridge for your platform — a small (~17 MB) native app that connects everything else. Full details on the download page.

First launch

Open the app. It looks at your hardware — how much RAM you have, whether there is a GPU or Apple Silicon — and recommends a starter model that actually fits your machine. One click downloads it.

The backends doing the heavy lifting (Ollama, LM Studio, Apple MLX) are installed and managed by the app itself. You never touch a terminal or install anything by hand.

Your first chat

Pick your model from the model picker and start typing. Every token is generated on your own machine — nothing is sent to a server, so it works offline too.

Response speed depends on your hardware: a small model on a recent laptop answers almost instantly, a big one on older hardware takes its time. That trade-off is yours to make.

Code with an agent

Switch to agent mode and choose a workspace folder. Inside that sandboxed workspace the coding agent can read and write files, run shell commands and execute your tests — fix a bug, scaffold a feature, or just explain what a repo does.

Because the model runs locally, your source code never leaves your disk. More on the agent and its tools on the features page.

Generate images

On Apple Silicon, image generation runs through MLX — SD-Turbo is a fast starter that renders in seconds, and photorealistic models are an optional download away. On NVIDIA GPUs the app manages a ComfyUI backend for you, with models like FLUX.1 Schnell.

It is your hardware, so there are no credits, no watermarks and no generation caps.

Try video (early)

Short text-to-video clips with open models like Wan 2.1 work today — but we label this early for a reason. Downloads start around 18 GB and generation wants serious hardware; on a mid-range machine expect long waits.

If your box can take it, it is genuinely fun. If not, skip it for now — chat, code and images are the mature parts.