AI Agent Deep Dive
A research report on AI agent source code, distributed as downloadable PDFs, alongside a small teaching project that demonstrates how a minimal Python agent's core loop, tool skills and CLI fit together.
This repository offers downloadable PDF reports that research AI agent source code in depth, including a chapter on a memory system and a newer analysis of the Hermes Agent's source code, alongside a minimal teaching Python agent project.
The teaching project keeps its core agent code in src/agt/agent.py and its CLI entry point in src/agt/cli.py, deliberately kept small so it is easy to learn from; it currently ships a replaceable fake LLM that echoes a user's input back as streamed text chunks, so that wiring in a real model later only requires replacing the LLM call layer rather than rewriting the whole agent.
It uses Poetry for dependency management and runs with poetry run agt; it is written in Python with no license stated in the material's facts.