tvytlx /ai-agent-deep-dive

AI Agent Deep Dive

★ 5,831⑂ 1,613👁 28⑃ 2● 3▤ 0
PythonCreated 2026-03-31
+0
Daily
5,831
stars
1,613
forks
28
watchers
2
prs
3
issues
0
releases

README

AI Agent Deep Dive

English | 简体中文


 
   
   
   
 


 Quick Links

PDF reports:
- New: Hermes Agent Source Code Deep Dive
- Claude Code: ai-agent-deep-dive-v2.1.pdf, including the new Chapter 8: Memory System.
- Claude Code: ai-agent-deep-dive-v2.pdf.

 Educational Agent Code

This repository also includes a minimal Python agent built for learning, demonstrating how to structure the core components of an AI agent.

 Where to Find the Code

- Agent core: src/agt/agent.py
- CLI entry point: src/agt/cli.py
- Learning materials: docs/

 Project Goals

This minimal agent is designed for educational use, with an emphasis on:

- Keeping the structure clear.
- Avoiding unnecessary engineering complexity.
- Keeping the core code small and easy to explore.
- Focusing on the agent loop, a Fake LLM interface, skill discovery, and the CLI skeleton.

 Running the Minimal Agent

The project uses Poetry to manage dependencies.

 1. Install Dependencies

bash
poetry install


 2. Run the Agent CLI

bash
poetry run agt "Hello"


 3. List Available Skills

bash
poetry run agt --skills-dir ./skills --list-skills


 Current Implementation

The current version is a minimal educational implementation. It does not yet connect to a real remote model API.

It includes a replaceable Fake LLM that returns a test response in streamed text chunks for each user input.

This keeps the model integration separate from the agent itself: connecting a real model later only requires replacing the LLM call layer, rather than rewriting the agent.

Top issues

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON