Repomix (Python Version)

Repomix is a tool designed to pack entire repositories into a single, AI-optimized file for use with LLMs (Claude, ChatGPT, Gemini).

Key capabilities include:

• AI-Optimized Formatting: Structures code for easy LLM processing.
• Token Counting: Uses tiktoken to provide counts per file and for the whole repo (configure via --token-count-encoding).
• Git-Aware: Automatically respects .gitignore, .repomixignore, and .ignore files.
• Security: Uses detect-secrets to prevent sensitive data inclusion.
• Compression: Multiple modes to reduce output size.
• Performance: Uses multiprocessing/threading for speed.
• Robustness: Automatically detects file encodings using chardet.

Repomix is designed to pack entire repositories into a single, AI-optimized file. Key features include:

• AI Optimization: Formats code for LLMs (Claude, ChatGPT, Gemini).
• Token Counting: Provides token counts per file and for the whole repo using tiktoken (configurable via --token-count-encoding).
• Git Awareness: Automatically respects .gitignore, .repomixignore, and .ignore files.
• Security: Built-in secret detection using detect-secrets.
• Code Compression: Multiple modes to reduce output size while preserving context.
• Multi-directory Support: Process multiple directories at once (e.g., repomix src lib tests).
• Performance: Uses multi-processing/threading for faster analysis.
• Encoding Support: Automatically detects file encodings using chardet.

Repomix supports advanced code compression using Tree-sitter, which allows you to reduce token usage while preserving essential code structures. This is useful for fitting large codebases into AI context windows.

Supported languages include: Python, JavaScript, TypeScript, Go, Java, C, C++, C#, Rust, Ruby, PHP, Swift, and CSS.

Repomix uses several methods to exclude files. When multiple methods are used, they are applied in the following priority order (highest to lowest):

1. ignore.custom_patterns in repomix.config.json
2. .repomixignore file
3. .ignore file (if ignore.use_dot_ignore is true)
4. .gitignore file (if ignore.use_gitignore is true)
5. Default patterns (if ignore.use_default_ignore is true)

All patterns use glob syntax similar to .gitignore.

Repomix can generate structured reference materials in the Claude Agent Skills format. This creates a directory structure under .claude/skills/<name>/ containing SKILL.md and a references/ folder with summaries, project structure, file contents, and tech stack information.

Use --skill-generate [name] to create the skill. You can specify a custom directory with --skill-output <path>.

# Generate skills with auto-detected name
repomix --skill-generate

# Generate skills with a custom name
repomix --skill-generate my-project

# Specify output directory directly
repomix --skill-output ./my-skills-dir

You can run Repomix in an isolated Docker environment. This is useful for avoiding local dependency conflicts and ensuring consistent results across different machines.

Build the image:

docker build -t repomix .

Run on the current directory:

docker run --rm -v "$(pwd)":/app repomix

Run with specific options (e.g., Markdown style and custom output):

docker run --rm -v "$(pwd)":/app repomix --style markdown --output custom-output.md

Run on a specific project directory:

docker run --rm -v "/path/to/your/project":/app repomix

You can run Repomix in an isolated environment using Docker. You must mount your project directory to the /app directory inside the container.

Build the image: docker build -t repomix .

Run in current directory: docker run --rm -v "$(pwd)":/app repomix

# Build the Docker image
docker build -t repomix .

# Run repomix in the current directory
docker run --rm -v "$(pwd)":/app repomix

# Run with specific options (e.g., markdown style and custom output)
docker run --rm -v "$(pwd)":/app repomix --style markdown --output custom-output.md

# Run on a specific project path
docker run --rm -v "/path/to/your/project":/app repomix

Repomix helps you estimate the size of your codebase in AI tokens.

Choosing an Encoding

Use --token-count-encoding to match your target model:

• o200k_base: Default, used for GPT-4o.
• cl100k_base: Used for GPT-3.5/4.

Visualizing Token Distribution

Use --token-count-tree to see a file tree with token counts. You can provide an optional threshold to only show files with at least $N$ tokens.

# Use GPT-3.5/4 encoding
repomix --token-count-encoding cl100k_base

# Show only files with 100 or more tokens
repomix --token-count-tree 100