Claude API · Python · 51 reference templates

The anatomy
of an agent.

Fifty-one small, runnable Python templates — one file per concept — organized like a field reference for agentic AI. Start from a single API call, work up through memory, tools, and planning, and end at multi-agent systems. Every template is heavily commented and built to be read, not just run.

Browse the index → View source on GitHub

Fig. 1 — the core agentic loop, and where each Core Architecture piece attaches to it

ROLESYSTEM
PROMPT ENGINELLM
BACKBONE LOOPREASON
→ PLAN ACTTOOL
USE OBSERVEENVIRON-
MENT

↩ feeds back into REASON via Memory/ · bounded every cycle by Execution_Loops/ · checked at every step by Safety_and_Control/

51Templates

10Concept modules

1API — Claude

0Framework dependencies

SETUPrun from repo root

# 1. install the one dependency
pip install -r requirements.txt

# 2. set your key — get one at platform.claude.com
export ANTHROPIC_API_KEY=your-key-here

# 3. run any template — start here
python3 Core_Architecture/basics/basic.py

The anatomyof an agent.

The anatomy
of an agent.