# Introducing gc - AI Powered git commit message generator Have you ever struggled to write a good Git commit message? I know I have! That’s why I created **gc**—a CLI tool powered by AI to help you write clear, meaningful commit messages without any hassle. ## Why did I build gc? ![](https://cdn.hashnode.com/res/hashnode/image/upload/v1732880959162/b73a83c3-92cd-41ad-82e9-2ae7d3b2d6ba.png align="center") The idea for **gc** came from a common developer struggle: writing good commit messages. During one of my collaborative projects, I noticed how unclear and inconsistent commit messages caused confusion. Some were too vague, like “fix stuff,” while others were unnecessarily long. This made it hard to understand the changes when revisiting the code or working with the team. I wanted to solve this problem by making commit messages simpler and smarter. With AI becoming more accessible, I realized it could help automate this task. The goal was to create a tool that saves time, ensures meaningful commit messages, and lets developers focus on coding instead of worrying about what to write in their Git commits. **gc** is my solution: an AI-powered CLI tool that generates clear and context-aware commit messages, but still gives you the flexibility to customize them. It’s all about making Git workflows smoother and reducing stress for developers. ## Usage **gc** is a globally installable Node.js package designed to make your Git workflow smoother. It’s easy to set up and works seamlessly with any Git repository. ### Installation To install **gc**, make sure you have Node.js (v16 or higher) and Git installed. Then, run: ```bash npm install -g @utsavdotdev/gc ``` That’s it! The tool is now ready to use. ### Commands * **Generate Commit Messages** ```bash gc new ``` Get AI-generated commit message suggestions and select or edit the one you like. * **Write Custom Commit Messages** ```bash gc new -c ``` Manually input your own commit message. * **Privacy Settings** ```bash gc opt-in gc opt-out ``` Choose whether to share anonymous data to help improve the tool. * **Help** ```bash gc --help ``` View a detailed list of commands and how to use them. ### Usage Using **gc** is as simple as these three steps: 1. **Stage Your Changes** Before running **gc**, add your changes to the staging area using: `git add .` 2. **Generate or Write a Commit Message** To let AI suggest commit messages, use: `gc new` ![](https://cdn.hashnode.com/res/hashnode/image/upload/v1732884215699/f63015a1-23b5-46b9-82d9-afaf65e11dc7.gif align="center") To write a custom commit message, use: `gc new -c` ![](https://cdn.hashnode.com/res/hashnode/image/upload/v1732884922903/7fa26b61-d8ec-4547-9fd5-0612181e43d5.gif align="center") 3. **Commit or Copy the Message** Once you have your message ready, you can: * Copy it to your clipboard for later use. * Open and edit it in your preferred CLI editor to make additional adjustments. * Commit it directly to your repository. ## Development Process Creating **gc** was an exciting journey that involved several steps to ensure it was useful, reliable, and user-friendly. Here’s a breakdown of the development process: ### Stage 1 - Research To kick off the project, I started by looking into existing solutions like [**ai-commit**](https://github.com/insulineru/ai-commit) and [**aicommits**](https://github.com/Nutlope/aicommits). Both are CLI tools designed to generate AI-powered commit messages. While they are useful, I noticed a few areas where they could be improved: * **Customization**: These tools lacked flexibility for users to edit messages easily during the workflow. * **Interactive Features**: I wanted a more interactive and user-friendly experience for selecting, editing, or copying messages. * **Privacy Options**: I felt it was important to give users clear control over anonymous data sharing(experimental) This research gave me a solid foundation and clear direction for building **gc**, focusing on making it smarter, more flexible, and better suited to developers’ needs. ### Stage 2 - Gathering Tech Stack Since the hackathon was organized by Modus, using their platform for building functions and APIs was a mandatory requirement. Modus integrated Models provided the foundation for generating context-aware commit messages, and integrating it became a key part of the project. For other aspects of the project, I carefully selected tools to ensure efficiency and scalability: * **Next.js**: To build a sleek and responsive website for showcasing **gc**. * **Node.js**: The backbone of the CLI tool for its performance and simplicity. * **Inquirer.js**: To create interactive prompts for a seamless user experience. * **Commander.js**: For command-line argument parsing and structuring the CLI commands. * **Clipboardy**: To enable copying generated commit messages directly to the clipboard. * **Simple-Git**: To interact with Git repositories programmatically. ### Stage 3 - Building Client Side The first step in building the client-side was creating the **CLI tool**, as it’s the core feature of **gc**. I focused on ensuring the tool was simple to use and provided a seamless experience for developers. Key features like AI-powered commit suggestions, custom messages, and clipboard integration were implemented during this phase. ![](https://cdn.hashnode.com/res/hashnode/image/upload/v1732934466686/37cf6f52-b090-468a-9813-48de43339f92.jpeg align="center") After finalizing the CLI tool, I moved on to designing the **landing page** using **Next.js** and **TailwindCSS**. The landing page includes a **mock terminal box**, where users can interact with predefined commands such as: * `gc info`: Displays information about the tool. * `gc features`: Lists the key features of **gc**. * `gc commands`: Shows a quick reference for all available CLI commands. * `cls` or `clear`: Clears the terminal screen. ### Stage 4 - Integrating Modus Integrating **Modus** was a crucial step in the development of **gc**. This was my first experience with **Hypermode** and **Modus**, so I started from scratch. I began by watching the collaborative YouTube video between **Hypermode** and **Hashnode**, which helped me understand the basics. Then, I dive into the documentation, exploring code snippets provided by **Hypermode**. Initially, it was overwhelming, and I felt completely lost. To clear my mind, I decided to focus on completing the client-side part of the project first. Once I felt refreshed, I returned to setting up **Modus**, following the documentation step by step. I was able to successfully set it up and generate a simple GraphQL API for a "hello message" function. Thankfully, GraphQL wasn’t new to me, as I had worked with it previously while experimenting with the **Hashnode** API. Next, I explored the models I would need. I initially tried using an external model by adding it to the `modus.json` file, but I encountered limitations due to free credits running out after testing. Fortunately, I discovered that **Hypermode** also hosts some models, which worked much better for my needs and provided valid responses.
💡
I used the Hypermode-hosted text generation model, meta-llama/Meta-Llama-3.1-8B-Instruct, to power the AI-driven commit message generation feature.
![](https://cdn.hashnode.com/res/hashnode/image/upload/v1732934623053/3de63af6-72b4-47eb-83af-adde437ac446.jpeg align="center") After that, I wrote a function that takes a prompt and the Git diff output, then returns an array of commit messages. To fine-tune this function, I tested it approximately 100-130 times, iterating on the prompt to get the most accurate results. However, I ran into a challenge when the service suddenly crashed, showing a 503 error. Fortunately, the **Modus** team resolved the issue within a couple of hours. Once everything was up and running, I fine-tuned the responses from the model and displayed them to the user in a clear and meaningful way. This integration of Modus into **gc** was a key step in making the AI-powered commit message feature come to life. ### Stage 5 - Deployment When it came to deploying **Modus**, I initially thought it would be a difficult process. However, once I aligned everything in the right way, it turned out to be much easier than I expected. For the **landing page**, I hosted it on **Vercel**, which was quick and simple, allowing it to be live right away. The **CLI tool** was published on the **npm package registry**, making it easy for others to install and use. As for the **Modus functions and APIs**, I had to link my GitHub repo to **Hypermode** using the command: ```bash hyp link ``` Once linked, it automatically updated the Modus app's endpoints with each commit and provided a production-ready API accessible with the **Hypermode API key**. This process was surprisingly easy and worked smoothly once everything was set up properly. ## Problem Encounter During the development of **gc**, I faced a few challenges that required some problem-solving and adjustments: 1. **Modus API Crashes**: While testing the integration with **Modus**, I encountered an issue where the API returned a 503 error. It turned out that this was caused by an overload in requests. Fortunately, the **Modus** team fixed the issue within a couple of hours, allowing me to continue working without major delays. 2. **External Model Limitations**: I initially tried to use an external AI model for generating commit messages by adding it to the `modus.json` file. However, after running out of free credits, I had to find an alternative. Luckily, **Hypermode** hosted models that worked well, and I was able to use those instead. 3. **Fine-Tuning AI Responses**: Getting the AI model to generate relevant commit messages based on Git diffs was challenging. I had to test the prompt multiple times—about 100-130 attempts—to refine it and ensure that the responses were meaningful. Even after that, I had to fine-tune the model’s outputs to match the expected results. ## Links %[https://youtu.be/UmwjiRI8pdk] * 🖥️ **GitHub Repository**: [utsavdotdev/gc](https://github.com/utsavdotdev/gc) * 🌐 **Website**: [gc](https://gc-cli.vercel.app/) ## Conclusion Building **gc** was a great experience, and I’m happy with how it turned out. It’s a tool that helps developers generate meaningful commit messages using AI, making workflows smoother and more efficient. A huge **thank you** to **Hashnode** and **Hypermode** for organizing this hackathon and providing the tools and support that made this project possible. 🚀