llm-agent

A minimal agentic coding assistant built in Java. It accepts a natural language prompt, calls an LLM via the OpenRouter API, and autonomously executes tool calls in a loop until the task is complete.

How it works

The agent runs a loop:

1. Sends the user prompt (plus conversation history) to the LLM
2. Receives a response — either a final answer or one or more tool calls
3. Executes the requested tools and feeds results back into the conversation
4. Repeats until the LLM returns a final answer (no more tool calls)

Tools

Tool	Description
Read	Read a file from disk
Write	Write content to a file
Bash	Execute a shell command

Requirements

• Java 25
• Maven
• An OpenRouter API key

Setup

export OPENROUTER_API_KEY=your_key_here

Optionally override the base URL (defaults to https://openrouter.ai/api/v1):

export OPENROUTER_BASE_URL=https://openrouter.ai/api/v1

Usage

./run.sh -p "your prompt here"

Example:

./run.sh -p "List all Java files in this directory and count the lines in each one"

Model

Defaults to anthropic/claude-haiku-4.5 via OpenRouter. Change the model in Main.java in the invokeLLMApi method.

Build

mvn package -Ddir=/tmp/llm-agent-build
java --enable-preview --enable-native-access=ALL-UNNAMED -jar /tmp/llm-agent-build/llm-agent.jar -p "your prompt"