Practical tutorials for local AI on the hardware you already own.
Every guide is written from hands-on experience, tested on the actual hardware it targets, and captures the traps that documentation never mentions. No theoretical walkthroughs — just what worked (and what didn’t).
Krea 2 Turbo on a GTX 1070
Running image generation on 8 GB Pascal GPUs. The 5 trap doors you’ll hit that aren’t documented anywhere — and one-step setup scripts for both Linux and Windows.
LiveGemma-4 Multimodal on Pascal
Google DeepMind’s latest multimodal model — text, images, audio, and video — running on a 2016 GPU. The three flags that make 6.86 GB fit in 8 GB VRAM, and analyzing a 100-frame storyboard.
LiveBonsai Ternary LLM on CPU
An 8-billion-parameter language model running on the CPU alone — no GPU, no CUDA, ~2 GB of RAM — via ternary (1.58-bit) quantization. Fast prefill (440 tok/s), honest generation numbers, and a full serving stack.
LiveLive Linguist v2
On-device easy-language caption simplifier running on Apple Silicon. Fine-tuned Qwen3 models validated against official easy-language registers in 20 languages.
LiveHermes Agent from Scratch
Setting up a personal AI agent that can use tools, write code, and run autonomously on your own infrastructure.
Coming soonBecause someone had to try it on the old GPU
The best AI tutorials assume you have an RTX 4090 or a cloud budget. Most people don’t. These guides are for the GTX 1070s, the M1 MacBooks, the GPU-less servers, the hand-me-down hardware that’s still perfectly capable.
Every guide includes the debugging that went into getting it working — the wrong flags, the silent crashes, the impossible-to-Google error messages. You shouldn’t have to repeat them.