I Built SlowCommit: A Local-First AI Git Commit Message Generator 🦥

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady Submission 🦥

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

I built SlowCommit 🦥, a local-first AI-powered Git commit message generator using Python and Ollama.

Writing clear commit messages is a repetitive task. SlowCommit helps developers save some effort by generating a concise Conventional Commit message from their staged Git changes.

It reads only the staged diff, applies best-effort redaction for certain common secret patterns, limits the prompt size, and asks a locally running AI model to suggest a message.

The developer reviews the suggestion and decides whether to use it. SlowCommit never commits automatically.

I tested SlowCommit locally on Windows using qwen2.5-coder:7b.

For a staged README update, it suggested:

docs: add local demo section to README Enter fullscreen mode Exit fullscreen mode This is a local command-line application rather than a deployed web app.

GitHub repository: https://github.com/mohit-saini-dev/slowcommit

The project includes a Python CLI, Git staged-diff handling, an Ollama API integration, basic secret-pattern redaction, bounded prompt construction, and automated tests.

I built SlowCommit in Python and integrated it with Git and Ollama.

The tool retrieves staged changes using git diff --cached, processes the diff with safeguards such as best-effort secret redaction and prompt-size limits, and sends it to the local Ollama API.

I tested it with the open-weight qwen2.5-coder:7b model. The project also includes tests for redaction, diff truncation, untrusted prompt framing, and commit-message cleanup. All six tests pass.

Open innovation makes it possible to build useful AI tools without requiring a hosted inference API.

With Ollama, users can run a supported model on their own computer and choose which model suits their hardware and needs. This gives developers more control over their workflow and can reduce dependence on paid cloud inference services.

SlowCommit is designed around that idea: local inference by default, a small and inspectable codebase, and human review before any commit is made.

Local processing does not guarantee that every secret will be detected, so users should still review staged changes and avoid staging sensitive information.

SlowCommit automates the first draft of a commit message, reducing repetitive writing effort while leaving the final decision to the developer.

I tested it on a real staged README change and received a usable suggestion. I haven't measured a reliable time saving yet, so I won't claim a specific number.