Ollama
A linear, one-concept-per-lesson path through Ollama, running open-source LLMs on your own machine instead of a cloud API, from your first local generate call to a fully offline RAG agent. 24 lessons · 3 tiers.
Prerequisites: Completion of LangChain & Agent Building through Lesson 11 (init_chat_model), where the model behind an agent was already shown to be swappable. Every other course in this repo calls Gemini over the network; this one builds the same kinds of agents, generate calls, tool-calling loops, and RAG, against a model running locally.
No GOOGLE_API_KEY needed at all: every lesson calls a model running locally via Ollama (llama3.2 by default), not a cloud API. Setup needs the Ollama application installed separately (not a Python package) from ollama.com/download, plus `ollama pull llama3.2` before starting. Lessons 8 and 21 also use a small local embedding model, nomic-embed-text.
Built on Ollama, the local model runtime this course teaches.
Beginner
Running and calling a local model: generate, chat, streaming, embeddings.
Intermediate
Structured output, tool calling, and swapping into LangChain or Pydantic AI.
Advanced
Hardware, context limits, concurrency, local RAG, and production concerns.