Local LLMs
Running large language models (LLMs) on your own machine with Ollama. CivicGrid's weekly mayor check uses a small local model on my homelab: it proposes changes, and a person approves every one.
Learning track · 13 lessons
Local LLMs with Ollama
From what a language model really does to running one as a checked, scheduled job. Each lesson explains one idea with an everyday example, has you try it with a small model on your own laptop, and practice on a made-up county whose right answers are known, then shows how CivicGrid's weekly mayor check uses the same idea, including where it goes wrong.
- 1What an LLM is, and isn'tStart · 20 min
- 2Your first local modelStart · 30 min
- 3Picking a model your machine can runStart · 25 min
- 4Talking to it from codeStart · 30 min
- 5Prompts that do one jobStart · 30 min
- 6Context windows: what the model can seeStart · 25 min
- 7Structured output: answers your code can trustStart · 30 min
- 8Checking its answers: catching made-up onesStart · 30 min
- 9Prompt injection: when the page gives ordersStart · 30 min
- 10Testing a model before you change itStart · 30 min
- 11Ask your own documents: embeddings and retrievalStart · 35 min
- 12Running it as a serviceStart · 30 min
- 13Local or hosted?Start · 25 min
Guides
A local LLM in production: the weekly mayor verifier
Why a small model on a home server, how it only proposes and never writes, and the regression cases every change has to pass.
Using a local model as a coding helper
A coding model on your own laptop for scripts: what it does well, where it invents functions that don't exist, and the checks before you run anything it wrote.
Tools I use
Ollama (opens in a new tab)
Download and run open models locally with one command, plus a simple local API.
Ollama model library (opens in a new tab)
Browse models by size, so you can pick one your hardware can actually run.
From www.kwamib.dev/resources/local-llms