How it works

Your machine is the whole stack.

There is no server tier. LU Labs detects what you already run, loads a model into your memory, and generates — start to finish — on-device.

01
Auto-detect your backends
LU Labs scans for the 12 runtimes it supports — Ollama, LM Studio, vLLM, llama.cpp and more — and lists what you can run.
02
Load a model into memory
Pick a model; it loads into your own RAM or VRAM. A green dot means it is live and ready — no download of your prompts anywhere.
03
Generate — on-device
Every token, image and frame is produced by your hardware. The result never round-trips through a cloud you do not control.

What leaves your machine?

The honest answer, side by side.

LU Labs
  • Nothing — not automatically, not ever
  • No prompts, no chat history, no files
  • No generated images or video
  • No accounts, emails or device IDs
  • No usage analytics or crash pings
Typical cloud AI
  • Every prompt you type
  • Files and images you attach
  • Conversation history for training
  • Account, IP and device fingerprint
  • Detailed usage telemetry
ON-DEVICE
100% local inference
The model runs where your data already lives. Air-gap it if you want.
NO TELEMETRY
Zero phone-home
No analytics SDK, no background reporting, no silent updates of your data.
NO KEYS
No API keys to leak
Local mode needs no cloud credentials, so there is nothing to steal.
OPEN SOURCE
Auditable code
Read exactly what the app does. The community can and does.
AUTO-CLEANUP
Processes die on close
The ComfyUI process is killed when you quit — no lingering daemons.
OFFLINE
Works with no internet
Once a model is on disk, pull the ethernet cable. It still runs.

Read the source. Trust the source.

LU Labs is open source — verify every claim on this page yourself.