AI and local models

Learn what a language model really does, run one on your own computer, make it answer from your own documents with sources, and call it from an API. Every lesson ends with the habits that keep your data and your machine safe.

What you will be able to do

By the end of this track you can:

• explain tokens, the context window, temperature and why a fluent answer can still be wrong;
• judge whether a model fits on your computer, and choose between a local and a cloud model;
• explain how retrieval-augmented generation (RAG) finds passages and cites them, and when it should say "I don't know";
• read and send an OpenAI-style chat request, and explain routing, failover and budgets in a gateway;
• explain how a model "uses tools", and what an MCP server and client do;
• keep secrets and personal data out of AI chats, spot prompt injection, and keep local AI servers private.

Who it is for

Students, testers, developers and curious home users. You need no maths and no machine-learning background. Some lessons show requests and JSON; if those are new, read What is an API? and HTTP basics first, or alongside.

The lessons

Take them in order. Each one builds on the one before.

Lesson 1How language models workTokens, next-token prediction, context, temperature and hallucination.→Lesson 2Run models on your own computerOllama, memory, quantisation, CPU or GPU, local or cloud.→Lesson 3Answers from your documents (RAG)Chunks, embeddings, a vector store, citations and honest refusals.→Lesson 4Gateways and AI APIsThe chat request, API keys, routing, failover, budgets and streaming.→Lesson 5Tools and MCPHow a model asks for a tool, and how MCP servers and clients fit.→Lesson 6Use AI safelySecrets, checking answers, prompt injection, localhost and licences.→LabAsk questions about your own documentsMarkdrop, LocalRAG and ApiWorkbench together, with tests.→

Apps that use this knowledge

App	What you do with it in this track
LocalRAG	Ask questions about your own PDF, text and Markdown files, with page citations, fully offline
DL Prism	One OpenAI-compatible endpoint on your machine that routes requests between local and cloud models
Markdrop	Turn Word files, PDFs and web pages into Markdown, from a desktop app, an extension or an MCP connector
Cloakpaste	Swap emails, keys and names for placeholders before you paste text into an AI chat
SignVox	An on-device recognition model, and a clear example of honest accuracy limits
ApiWorkbench	Call an AI HTTP API with saved requests and tests

How long it takes

Each lesson takes 15 to 25 minutes, including its quiz. The lab takes 45 to 60 minutes, plus download time the first time you install the models. Allow about three hours for the whole track, spread over as many sittings as you like.

Next steps

• Start with How language models work
• See your scores on the Quizzes page
• Browse the other learning tracks