Local AI Local AI Foundations

Local AI Landscape

Map the self-hosted AI stack before you pick your first tool

47 of 66

Why run AI locally?

Cloud AI is fast and capable, but it sends your code, prompts, and data to someone else's server. Local AI keeps everything on your machine: private, offline, and with no per-token bill. It is also the cheapest way to experiment aggressively.

The three local stacks

StackBest forEntry point
LM StudioGUI-first chat, quick experiments, OpenAI-compatible serverDownload app
OllamaTerminal-first workflows, automation, ModelfilesOne-line install
Image generationCreative assets, thumbnails, dataset generationComfyUI

Key concepts

  • Model weights are the trained files (several GB each).
  • Quantization shrinks weights to trade a little quality for a lot of speed.
  • VRAM on your GPU decides how large a model fits without slowing down.
  • CPU offload lets slower machines run big models using system RAM.

Choose your first stack

  • Want a chat UI today? Start with LM Studio.
  • Want scripts and terminal control? Start with Ollama.
  • Want images? Start with ComfyUI.

You can mix them later. LM Studio and Ollama can both expose the same local model through an API.

Working out which model to run this on? See The Codex. Packaging it as a reusable skill? See The Armory.