BundleRepo

BundleRepo is a beta tool designed to clone and pack a local or remote (GitHub only for now) Git repository into a comprehensive XML file. The packed XML includes detailed metadata about each file, such as the size in bytes and the number of lines, making it suitable for large language model (LLM) consumption, code analysis, and repository review.

XML was chosen for the file output format since it is very well structured and LLM models can easily parse it (better than a plain-text dump).

It is inspired by Repopack which is a great tool, but is written in TypeScript and needs a Node.js environment to run. Eventually this project will produce binaries and not need Rust installed to run.

The generated XML metadata and structure are inspired by the output of Repopack (a lot of the header text was taken from there), with enhancements that include additional file attributes, instructions for the LLM and a more robust structure. At this time xml output is the only supported output format, however future versions may include additional formats.

XML was chosen as the default output format since it is very well structured and LLM models can easily parse it (better than a plain-text dump - see this link from Anthropic as to why XML is a superior format for feeding context and instructions into an LLM).

BundleRepo Version 0.1.0, © 2024 Grant Ramsay <seapagan@gmail.com>

Pack a local or remote Git Repository to XML for LLM Consumption.

-> Found a git repository in the current directory: '/home/seapagan/data/work/own/bundle-repo' (branch: main)
-> Successfully wrote XML to packed-repo.xml

Summary:
     Total Files processed:  11
 Total output size (bytes):  47906
      Token count (GPT-4o):  11344

• Compatibility
• Features
• Usage
  • Installation
  • Running the Tool
    • Specify the branch for a remote Git repository
  • Output
    • Output to File
    • Output to stdout
    • Copy to Clipboard
    • Add line numbers
  • Choose Model for Token Count
  • GitHub Token
• Command Line Options
• Configuration File
• Ignored Files
• Planned Improvements
• XML Layout
• Beta Status
• Acknowledgements
• License

Compatibility

The tool is designed and tested to work on Linux, MacOS, and Windows (Windows 10 and 11 tested).

Features

• Clone Git Repositories: Supports cloning both public and private repositories (with token support). Only supports https URLs at this time.

• File Scanning: Automatically scans the repository and adds all files to the output, excluding standard ignored files (e.g. .gitignore, LICENSE, etc).

  Any file listed in a .gitignore file will be excluded from the output and metadata.

  Binary file content will always be excluded, though they will be listed in the <repository_structure> node and a <file> node will be created in the XML to show that the file was excluded and why.

  See Ignored Files for a full list of excluded files.

• Metadata Extraction: For each file, the XML output includes:

  • path: the file path relative to the repository root
  • size: file size in bytes
  • lines: number of lines in the file
  • Raw file content (not escaped)

• Token Count: Calculates the number of tokens in the final XML file, based on the specified model (default is GPT-4o). Only OpenAI models are supported at this time, though I may add support for others in the future.

• XML Output: Generates an XML file (packed-repo.xml) that contains the entire repository structure and file details.

• Global and local configuration files: Allows you to set default values globally and override them on a per-project basis. All settings can be further overridden by command line options.

This tool is currently under active development, and more features will be implemented quickly. Please star this repository to stay updated on new releases and features.

Usage

This will be available as a binary download in the future, but for now, you can build it from source or install from crates.io. You will need to have Rust installed on your system to build the project.

Installation

1. Clone the project and install dependencies.

   • From crates.io:

     cargo install bundle_repo
     

   • From source:

     git clone https://github.com/seapagan/bundle-repo.git
     cd bundle-repo
     cargo build --release
     

     Move the binary to a directory in your PATH:

     eg for Linux or MacOS:

     sudo mv ./target/release/bundlerepo /usr/local/bin
     

Running the Tool

Use the GitHub short form:

bundlerepo user_name/repo_name

Use the full URL:

bundlerepo https://github.com/user_name/repo_name

Or use the current directory (if it is a git repository):

bundlerepo

Only the https protocol is supported at this time. The tool will not yet work with ssh URLs (ie not git@github.com:seapagan/bundle-repo.git)

The tool will actually bundle any files in the current directory (unless they are in the hard-coded ignore list). This can probably be useful for bundling any related files that you wish to feed to an AI. However, you may need to edit the <purpose> and <instructions> nodes in the output XML. I may add a flag to make this easier in the future (--not-code or something).

However, it still needs to be an actual git repository or the code will exit. I may add a flag to allow non-git repositories in the future.

Specify the branch for a remote Git repository

If you want to specify a branch for a remote repository you can do so using the --branch or -b flag:

bundlerepo user_name/repo_name --branch my_branch

Without this flag, the default branch will be used, which is usually main or master.

The --branch option only works for remote repositories. It has no effect when bundling a local repository. If you want to bundle a local repository with a specific branch you will need to check out that branch before running the tool.

Output

Output to File

This is the default operation of the tool, the XML output will be written to packed-repo.xml, which contains the hierarchical structure and metadata of the repository files. This can then be passed to an LLM model for analysis (for example, attach the output file to a ChatGPT or Claude prompt).

The filename can be changed using the --file or -f flag:

bundlerepo user_name/repo_name --file my-repo.xml

The output file will be written to the current directory unless a path is specified:

bundlerepo user_name/repo_name --file /path/to/output.xml

Output to stdout

You can output the XML to the terminal by using the --stdout or -s flag:

bundlerepo user_name/repo_name --stdout

This will print the XML output to the terminal, which can then be redirected to a file or piped to another application.

In this case, the --file flag is ignored and no file is written to disk.

Copy to Clipboard

You can copy the XML output to the clipboard by using the --clipboard or -c flag:

bundlerepo user_name/repo_name --clipboard

This will copy the XML output to the clipboard, which can then be pasted into another application or file, or indeed directly into an LLM prompt. Note that it is likely to be a large amount of text, so ensure your clipboard can handle it.

In this case, the --file flag is ignored and no file is written to disk.

Add line numbers

If you want to add line numbers to the output, you can use the --lnumbers or -l flag:

bundlerepo user_name/repo_name --lnumbers

This will add line numbers physically to each line in the output, which can be useful for debugging or analysis. Note that this will increase the token count of the output, so be aware of that when using it. Extra info for the LLM will be added to the <instructions> node to explain the line numbers.

Choose Model for Token Count

After generating the xml file, the tool gives a count of the number of tokens in the file, to give you an idea of context usage and costs. By default it calculates the number of tokens for the GPT-4o model, but you can specify another model using the --model or -m flag:

bundlerepo user_name/repo_name --model gpt3.5

Valid models are gpt4o, gpt4, gpt3.5, gpt3 and gpt2. It is important to use the correct model, as the token count is vastly different between the 3 and 4 series models.

Only OpenAI models are supported at this time, since the code uses the tiktoken library from OpenAI to count the tokens. I may add support for other models in the future, if I can find a decent library that supports them.

Currently, the count returned by this tool is identical to that returned by their web app.

GitHub Token

For private repositories, or to bypass usage restrictions, you can provide a GitHub token to access the repository. You can create a token by following the instructions here.

Once you have the token, you can pass it to the tool using the --token flag:

bundlerepo user_name/repo_name --token YOUR_GITHUB_TOKEN

Passing a token is totally optional if you are only using public repositories.

Command Line Options

The full list of command line options can be seen by running with the --help flag:

Pack a local or remote Git Repository to XML for LLM Consumption.

Usage: bundlerepo [OPTIONS] [REPO]

Arguments:
  [REPO]  GitHub repository to clone (e.g. 'user/repo' or full GitHub URL). If not provided, the current directory will be searched for a Git repository.

Options:
  -b, --branch <BRANCH>     Specify a branch to checkout for remote repositories
  -f, --file <OUTPUT_FILE>  Filename to save the bundle as. [default: packed-repo.xml]
  -s, --stdout              Output the XML directly to stdout without creating a file.
  -m, --model <MODEL>       Model to use for tokenization. Supported models: 'gpt4o', 'gpt4', 'gpt3.5', 'gpt3', 'gpt2' [default: gpt4o]
  -c, --clipboard           Copy the XML to the clipboard after creating it.
  -l, --lnumbers            Add line numbers to each code file in the output.
  -t, --token <TOKEN>       GitHub personal access token (required for private repos and to pass rate limits)
  -e, --extend-exclude <PATTERN>  Additional file pattern to exclude (can be specified multiple times)
  -x, --exclude <PATTERN>   File pattern to exclude, replacing the default ignore list (can be specified multiple times)
  -V, --version             Print version information and exit
  -h, --help                Print help

Configuration File

The tool supports two configuration files:

• Global config at ~/.config/bundlerepo/config.toml
• Local config at .bundlerepo.toml in your current directory

This allows you to set default values globally and override them on a per-project basis. All settings can be further overridden by command line options.

The configuration files use TOML format. Here's an example configuration:

# ~/.config/bundlerepo/config.toml or .bundlerepo.toml
output_file = "my-default-output.xml"
model = "gpt3.5"
stdout = false
clipboard = false
line_numbers = true
token = "your-github-token"
extend_exclude = ["*.md", "*.txt", "docs/*"]  # Additional patterns to exclude

All settings are optional. Settings are applied in the following order of precedence (highest to lowest):

1. Command line options
2. Local config file (.bundlerepo.toml)
3. Global config file (~/.config/bundlerepo/config.toml)
4. Built-in defaults

Available configuration options:

• output_file: Default output filename (default: "packed-repo.xml")
• model: Default model for token counting (default: "gpt4o")
• stdout: Whether to output to stdout by default (default: false)
• clipboard: Whether to copy to clipboard by default (default: false)
• line_numbers: Whether to add line numbers by default (default: false)
• token: Your GitHub personal access token (default: none)
• extend_exclude: Additional file patterns to exclude (default: none)
• exclude: File patterns to exclude, replacing the default ignore list (default: none)

The extend_exclude and exclude options can be specified either by using multiple -e or -x flags on the command line:

bundlerepo user/repo -e "*.md" -e "*.txt" -e "docs/*"
bundlerepo user/repo -x "*.exe" -x "*.dll" -x "node_modules/*"

Or as arrays in the TOML configuration file:

extend_exclude = ["*.md", "*.txt", "docs/*"]
exclude = ["*.exe", "*.dll", "node_modules/*"]

The extend_exclude patterns will be added to the default ignore list, while the exclude patterns will replace the default ignore list entirely.

Important: When the exclude option is used (either via command line or config file), both the default ignore list and any extend_exclude patterns are completely ignored. The exclude patterns become the only ignore rules in effect.

Note: The extend_exclude option is useful for excluding additional files that aren't in the default ignore list but that you don't want to include in your XML output. The exclude option gives you complete control over what files are ignored, replacing the built-in ignore list. Both options can help reduce token usage and remove irrelevant files from the LLM context.

Storing your GitHub token in the configuration file can be more convenient than passing it via command line, especially if you frequently work with private repositories. Just be sure to keep your configuration file secure.

Ignored Files

The tool will ignore the following files by default and (except for binary, see below) they will not be listed anywhere in the XML output:

• ANY Binary File. If you have a binary file in your repository, it will be listed in the XML output, but the content will be excluded.
• .gitignore
• any file listed in a .gitignore file
• .git folder and it's contents
• .github folder and it's contents
• Python requirements files (requirements.txt, requirements-dev.txt, etc)
• Lockfiles - any file ending in .lock
• renovate.json
• license files (e.g. LICENSE, LICENSE.md, etc)

This list is hard-coded (and to be honest is tuned to my current workflow) and cannot be changed at this time. However, that will be changed once the configuration file functionality is added.

I'm very open to adding other files that should be ignored by default, If you have a suggestion, please open a PR or an Issue on GitHub. For example, tool configuration files (eslintrc, prettierrc, etc), which are not needed by an LLM and just take up token space.

If there is demand, I may add a flag to allow the user to bypass this list and include all files. However, binary files will always be excluded as they don't fit well in XML.

Planned Improvements

You can find planned improvements and known issues etc in the TODO.md file.

XML Layout

The generated packed-repo.xml follows a structured format that can be easily understood by an LLM. Below is an example layout with explanations for each tag:

<repository>
  <file_summary>
    <!-- Metadata describing the purpose and file structure of the packed repository -->
    <!-- It also contains some instructions to help the LLM properly decode and understand the data -->
  </file_summary>

  <repository_structure>
    <summary>
      <!-- A brief summary of the folder structure in the repository -->
    </summary>
    <folder name="src">
      <!-- Folders contain nested folders and files -->
      <file path="main.rs">
        <!-- Files are listed by path relative to the repository root -->
      </file>
    </folder>
  </repository_structure>

  <repository_files>
    <summary>
      <!-- A summary of the files and their contents -->
    </summary>
    <file path="src/main.rs" size="1474" lines="53">
      <!-- For each file, the path, size in bytes, and number of lines are provided -->
      <!-- Full file contents are included here -->
    </file>
  </repository_files>
</repository>

Beta Status

This tool is currently in beta. While the core functionality works, there may be edge cases or features yet to be fully refined. Feedback and contributions are welcome to improve and stabilize the tool.

There is a pressing need for a test suite to ensure the tool works as expected in a variety of scenarios. This is a priority for the next release.

Acknowledgements

Bundle Repo is a rewrite of the original Repopack project, though none of the source code was used or even looked at (the output file header however was heavily borrowed from). The idea was to create a similar tool from scratch, with a few enhancements and improvements. It's also part of my journey to learn Rust and build useful tools for all.

License

This project is licensed under the MIT License.

The MIT License (MIT)
Copyright (c) 2024 Grant Ramsay

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE
OR OTHER DEALINGS IN THE SOFTWARE.