claude-code-from-scratch helps you understand how a coding agent works by rebuilding the core ideas in a simple way.
This project is made for people who want to:
- see how an AI coding tool is built
- follow a clear step-by-step guide
- run the result on Windows
- learn the basic flow behind Claude Code style tools
The repo uses TypeScript and Python in a small, readable code base. It focuses on the main parts that matter most:
- chat input
- tool use
- file handling
- task planning
- code editing flow
- agent loop logic
Use this release page to download and run the app on Windows:
After you open the page:
- Find the latest release at the top
- Open the Assets section
- Download the Windows file
- Run the file after it finishes downloading
If Windows shows a security prompt:
- Click More info
- Click Run anyway if you trust the file
- Wait for the app to start
If the download comes as a zip file:
- Right-click the zip file
- Choose Extract All
- Open the folder
- Start the app file inside
This app runs best on:
- Windows 10 or Windows 11
- 8 GB RAM or more
- 2 GB free disk space
- A stable internet connection
- A modern browser for some setup steps
For a smooth start, keep these ready:
- your Anthropic API key
- a text editor if you want to inspect files
- permission to run downloaded apps on your PC
After you install the app:
- Open the program
- Enter your API key if the app asks for it
- Choose a task or open a project folder
- Type what you want the coding agent to do
- Let the app plan, edit, and update files
- Review the changes before you save or run them
Good first tasks:
- ask it to explain a folder
- ask it to fix a small bug
- ask it to create a new file
- ask it to rename a function
- ask it to add a simple feature
The project breaks the coding agent into clear parts.
You type a request in plain language.
The agent breaks the request into steps.
The app reads files in the project folder.
The agent changes code where needed.
The app reviews the result and decides what to do next.
The agent repeats the cycle until the task is done.
This makes the system easier to understand than a large closed-source code base.
The code is organized so each part has a clear job.
Common folders and files may include:
src/for the main source codeapp/for the user-facing partsagent/for the agent loop and logictools/for file and task helpersdocs/for the step-by-step guidescripts/for setup and local run commands
You can use the guide chapters to move from one part to the next without getting lost.
This repo helps you learn:
- how a coding agent thinks through a task
- how an AI tool reads and writes files
- how tool calls work in a real app
- how prompts shape model behavior
- how to keep a simple agent loop clean
- how to build a working tool from small parts
If you want to understand Claude Code style systems, this project gives you a clear path.
Use this app when you want to:
- study how coding agents work
- test small code changes on Windows
- explore agent design without a large code base
- build your own version from a simple starting point
- follow a tutorial that stays close to real use
Before you start the app, check these items:
- you downloaded the latest release
- you extracted the files if needed
- you have your API key ready
- your internet connection works
- Windows Defender or your security tool allows the app to run
Then:
- Open the app
- Enter the setup details
- Load a folder with code
- Send a short test prompt
- Confirm the agent can read and edit files
Many AI apps need an API key to talk to the model.
To set it up:
- Get your key from your Anthropic account
- Open the app settings or setup screen
- Paste the key in the field
- Save the settings
- Restart the app if needed
Keep your key private. Do not share it in chat or in public files.
Try these steps:
- Download the release again
- Check that the file finished downloading
- Unzip the file if it came in an archive
- Run the app as administrator
- Turn off blocked file settings in Windows if needed
- Make sure your antivirus did not remove the file
- Check that your PC meets the basic system needs
If the app opens but does not respond:
- confirm your API key is correct
- check your internet connection
- open the app again
- try a short prompt first
The repo includes 11 chapters that build the project piece by piece.
Typical chapter flow:
- project setup
- app shell
- prompt input
- file reading
- file writing
- tool calls
- agent loop
- task planning
- error handling
- UI flow
- final integration
Follow the chapters in order to understand how each part fits together.
If you need the latest build or a fresh copy, use the release page here:
This project touches on:
- ai
- ai-agent
- anthropic
- build-from-scratch
- claude
- claude-code
- coding-agent
- llm
- tutorial
- typescript
When you run the app, use a test project folder first.
Good practice:
- create a small sample folder
- keep a backup copy
- test on a toy project before real work
- review each file change before you trust it
- use short prompts at first
- Open the release page
- Download the Windows file
- Run the file
- Open the app
- Add your API key
- Load a project folder
- Type your first task
- Review the result