Local AI, explained

Run a real AI model with zero internet

Tweets hype "free offline AI" constantly. The physics behind it is simple: model weights are just numbers, and if they fit in your RAM, they run on your machine. Here's exactly how the math works.

Drag to rotate the memory board. Each block = 1 GB of weights loaded.

Will it fit? RAM calculator

4.0 GBWeights
5.5 GBTotal RAM
~28 t/sEst. speed*
Fits comfortably on a 16 GB laptop.

*Speed estimate assumes a modern laptop with ~100 GB/s memory bandwidth. Generation speed ≈ bandwidth ÷ bytes read per token, so smaller quantized models are faster.

The worked example behind the hype

A model with 8 billion parameters stored at full 16-bit precision needs 8B × 2 bytes = 16 GB just for weights. Quantize to 4 bits and it becomes 8B × 0.5 bytes = 4 GB — small enough for an ordinary laptop, with quality loss usually under a few percent on benchmarks.

1
Download once

Tools like Ollama, LM Studio or llama.cpp pull a quantized GGUF file. After that, no network is touched.

2
Load into RAM

Weights map into memory. Apple Silicon's unified memory is why MacBooks punch above their weight here.

3
Generate locally

Every token is pure matrix math on your CPU/GPU. Zero cost per message, and your code never leaves the machine.

Popular local models (Q4 quantized)

ModelParamsRAM neededGood for
Qwen2.5-Coder 7B7B~5 GBCoding agents, autocomplete
Llama 3.1 8B8B~6 GBGeneral chat, summaries
Mistral Small 24B24B~15 GBStronger reasoning
Llama 3.3 70B70B~40 GBNear-frontier quality, needs a beefy Mac/workstation

Reality check: viral posts about brand-new "free local models that build whole apps" are often exaggerated. Local 7–8B models are genuinely useful for boilerplate, refactors and offline privacy — but they still trail frontier hosted models on hard, multi-step engineering tasks. Verify claims by running the model yourself; that's the whole point of local AI.

Enjoy this tool? Build your own with Super