Run AI on Your Own Machine

The parts, in order

1. 01 Pick Hardware That Can Actually Run Models Work out what your machine can hold using published Ollama tag sizes, the bits-per-weight arithmetic, and the GPU list Ollama actually documents. Beginner ~8 min Verified July 31, 2026
2. 02 Install Ollama and Run Your First Local Model Install Ollama on macOS, Linux, Windows or Docker, pull a first model, learn the ten commands you need, and find the logs when it refuses to start. Beginner ~8 min Verified July 31, 2026
3. 03 Run Laguna XS.2 Offline With Ollama: Setup Guide Install Ollama, pull the Laguna XS.2 build that fits your laptop, fix the 4096-token default, and wire it into your editor with no API key. Intermediate ~7 min Verified July 26, 2026
4. 04 Choose a Quantization Without Guessing Decode Q4_K_M, IQ2_XS and MXFP4 using llama.cpp's own measured bits-per-weight table, and see what the compression actually costs you. Intermediate ~9 min Verified July 31, 2026
5. 05 Run Stable Diffusion on Your Own PC: A First-Timer Setup ComfyUI's own docs, step by step: install the desktop app, put the 2.13 GB v1.5 checkpoint in the right folder, and generate offline without a terminal. Beginner ~9 min Verified July 26, 2026
6. 06 Wire a Local Model Into Your Editor Point Zed, JetBrains, Continue or OpenCode at http://localhost:11434, split one model into roles, and handle the doc drift you will hit on the way. Intermediate ~9 min Verified July 31, 2026
7. 07 Serve a Local Model Over an OpenAI-Compatible API Point any OpenAI SDK at http://localhost:11434/v1/, learn the request fields Ollama drops, and understand why the API key it asks for is ignored. Intermediate ~9 min Verified July 31, 2026
8. 08 Hermes Agent Tutorial: Self-Improving AI Setup Install and configure Hermes Agent so it remembers your workflows, builds reusable skills, and gets better at your tasks over time. Intermediate ~9 min Verified May 2, 2026
9. 09 Build Local RAG Over Your Own Files Index your documents with Ollama's /api/embed and Chroma, then retrieve offline. Covers the 2K embedding window trap and the distance-metric mismatch. Intermediate ~10 min Verified July 31, 2026
10. 10 When Local Is the Wrong Answer Four places local models lose on published numbers: context window, single-stream throughput, the model tier you cannot load, and your own unbilled setup time. Beginner ~8 min Verified July 31, 2026

Prerequisites are advisory. Every part is written to work on its own, so start wherever your problem is.