From attention mechanisms to threat modeling: the full agent stack

What it does Companion code for the O’Reilly book AI Agents – The Definitive Guide. The repository contains twelve chapter folders of Jupyter notebooks covering agent architectures, planning patterns, production deployment, evaluation frameworks, and threat modeling. Each notebook includes a Colab launch button, and a utils folder holds shared helper code.

The interesting bit Most agent repositories stop at a flashy ReAct demo; this one keeps going into contracts, tool governance, operational observation, and cost optimization. It is essentially a computer science textbook that happens to be executable.

Key highlights

• Twelve chapters spanning the full lifecycle, from foundational LLM mechanics to multi-agent systems and persistent memory design
• Production-oriented topics: deployment patterns, evaluation frameworks, secure execution, and test-time compute efficiency
• Notebooks are organized by chapter (CH01–CH12) with descriptive, topic-driven filenames
• Includes supplementary resources and a utils package for shared functions
• GPU may be required for some notebooks

Caveats

• The repo is companion material for a commercial book, so narrative context and explanations likely live on the printed page rather than inside the notebooks
• Some notebooks require a GPU to run

Verdict Worth bookmarking if you are building agents for production and want runnable reference material across the whole stack. Skip it if you are looking for a drop-in framework or copy-paste snippets that work without the book.