Agents From Scratch

agents_from_scratch is a small Python project for learning agent design from first principles. It focuses on a clear baseline: tool registration, simple tool routing, and predictable behavior.

Project goals

• Keep the core agent logic small and easy to inspect.
• Provide runnable examples for experimentation.
• Offer a clean package foundation that can grow into more advanced agents.

What is included

• SimpleAgent: a minimal text agent with pluggable tools.
• Built-in sample tools:
  • upper for text transformation.
  • math for basic arithmetic evaluation.
• CLI demo entrypoint: agents-from-scratch-demo.
• Example script and tests to show expected behavior.

Quick start

uv sync
uv run agents-from-scratch-demo

Try prompts like:

• help
• upper: hello world
• math: (2 + 3) * 4
• what can you do?

Repository layout

• agents_from_scratch/: package source code
• examples/: runnable examples
• tests/: package tests

Package name

The distribution name is agents-from-scratch, and the import package is agents_from_scratch.