Build Your Own Claude Code - Java

A from-scratch implementation of an AI coding assistant in Java. The assistant uses an LLM via the OpenRouter API to understand natural language prompts and autonomously complete multi-step coding tasks through an agent loop.

---

How It Works

The program implements a full agent loop:

1. Send the user's prompt to the LLM along with a set of available tools
2. If the model requests a tool call, execute it and feed the result back
3. Repeat until the model produces a final text response with no tool calls
4. Print the final response and exit This mirrors how real AI coding assistants (like Claude Code) operate — the model reasons step by step, using tools to explore and modify the codebase until the task is complete.

---

Tools

Tool	Description
Read	Reads and returns the contents of a file
Write	Writes (or overwrites) content to a file
Bash	Executes a shell command and returns stdout + stderr

---

Tech Stack

• Java (via Maven)
• openai-java SDK — OpenAI-compatible client used to call OpenRouter
• OpenRouter — API gateway providing access to anthropic/claude-haiku-4.5
• Jackson — JSON parsing for tool call arguments

---

Setup

Prerequisites

• Java 25
• Maven
• An OpenRouter API key

Environment Variables

export OPENROUTER_API_KEY=your_api_key_here
export OPENROUTER_BASE_URL=https://openrouter.ai/api/v1  # optional, this is the default

Run

./your_program.sh -p "Your prompt here"

Examples

# Read a file
./your_program.sh -p "What is the content of README.md?"
 
# Write a file
./your_program.sh -p "Read README.md and create the required file. Reply with 'Created the file'"
 
# Multi-step task using the agent loop
./your_program.sh -p "Use README.md to determine the chemical expiry period in months. Number only."
 
# Delete a file
./your_program.sh -p "Delete the old readme file."

---

Project Structure

src/
└── main/
    └── java/
        └── Main.java        # Entry point — agent loop, tool definitions, tool execution
pom.xml                      # Maven build config
your_program.sh              # Run script (compiles + executes)
codecrafters.yml             # CodeCrafters config (Java 25 buildpack)

---

Stages Completed

This implementation covers the following CodeCrafters challenge stages:

1. Send a prompt — Basic API call with user message
2. Advertise tools — Add tool specs to the request
3. Execute Read tool — Detect and handle a single tool call
4. Agent loop — Multi-turn conversation with persistent message history
5. Write tool — Create and overwrite files
6. Bash tool — Execute shell commands

---