gpt-context

What does it do? It generates a text representation of a git repository that can be used with LLM's (Large Language Models) such as Chat GPT! Check out the demo below.

gpt-context-demo.mp4

How is this useful? Well, you can use it to provide ChatGPT and other LLM's with structured data describing the layout of git repositories and the code they contain. That means you can use it with Chat GPT without the need for a paid for API key, or you can ask other LLM's for info about the code you want to summarize. Check it out with a real world example and Chat GPT.

chat-gpt-demo.mp4

Description/Purpose

The purpose of this shell script is to allow developers to bundle up git repositories of code into a format that is easy for Chat GPT to understand and parse. This enables a developer to upload their repository in plain text and then ask GPT questions about the code, structure, tooling, etc in the repository. There is no need for an OpenAI key to do this and the output file can be used with any LLM, it isn't restricted to Chat GPT.

gpt-context is git aware, it will read and parse a .gitignore file and deliberately ignore those files if they exist in the git repository and exclude them from the generated context.

Compatability

gpt-context is compatible with any POSIX compliant shell for portability. It can be used with macOS, Linux and WSL (Windows Subsystem for Linux)

Prerequisites

These are almost certainly available on your system out of the box.

• awk
• sed
• find

Setup

It's necessary to copy the script somewhere that is on your path as defined in your shell of choice.

cd to the directory that contains gpt-context e.g. cd ~/Downloads, make the script executable and then move it to somewhere on your path.

mkdir ~/bin
chmod +x gpt-context
mv gpt-context ~/bin/gpt-context

If ~/bin is not on your path, you can update your path by adding the following line to your shell profile.

export PATH="$HOME/bin:$PATH"

Now, after sourcing your shell profile, e.g. source ~/.zshrc or source ~/.bashrc the gpt-context command can be executed from any directory.

Usage

Usage: $0 [--help] [--version] [--input PATH] [--output OUTPUT]
  Defaults input and output to the current directory if not provided.
  --help         Show this help message
  --version      Show version information
  --input  PATH  Git repository path
  --output OUTPUT The path for the output (defaults to gpt_context.txt)

Example commands:

# Change directory to a git repository, I'm using my AlertR project from Github
~ cd AlertR
~ gpt-context
Generating GPT Context for GIT repository: ./
GPT Context saved to: gpt_context.txt

Alternatively, if you do not wish to have your current working directory as the root of a git repository a path can be provided to gpt-context. gpt-context will then create a gpt_context.txt file in your current working directory based on the path provided to the --input argument. The --output argument is used to specify an output file name that will be written to the current directory.

# change directory to home
cd ~
~ gpt-context --input ./personal-projects/AlertR --output alertr-context.txt
Generating GPT Context for GIT repository: ./personal-projects/AlertR
GPT Context saved to: /home/USER/alertr-context.txt

Once you have a generated context file it can be uploaded to Chat GPT or fed into any other LLM as part of the prompt to provide the LLM context. After this, the LLM can be asked questions about your code.