Stop Writing Changelogs by Hand. Let AI Do It Instead.

If you're a developer, you know the pain. You've just pushed a huge release with dozens of meaningful changes, and now comes the dreaded task: writing the changelog.

Did I remember everything from the past two weeks? Was that bug fix "critical" or just "minor"? Do users care about the internal refactoring or just what broke/got added?

For too many teams, the changelog workflow is: git log --oneline → scroll through 200 commits → cry a little → write something that's probably wrong.

The Problem with Manual Changelogs

Manual changelog writing has a few fundamental problems:

1. You forget things - Two weeks after the fact, you can't remember why you changed that one function from getUser() to fetchUser(). Was it important? Did it affect users?

2. Inconsistent formatting - Sometimes you write "Fixed:", sometimes "Fix:", sometimes you forget entirely. Your README.md and CHANGELOG.md don't match.

3. Wrong audience - As developers, we tend to write changelogs for other developers. But most users just want to know "what changed that affects me?" without the technical jargon.

4. Time sink - For non-trivial projects, writing a good changelog can take 30-60 minutes. Every. Single. Release.

What If AI Could Do This?

I started wondering: what if an AI could analyze my git commits and generate a reasonable changelog automatically?

The key insight was that git commit messages (when written well) already contain most of the information needed:

• What was changed
• Why it was changed
• Whether it was a fix, feature, or breaking change
• Context about what areas of the codebase were affected

Modern language models are surprisingly good at understanding natural language patterns and categorizing technical changes. Could they parse commit messages like a senior developer reviewing pull requests?

Enter KittyLog

KittyLog is the answer to this problem. It's a GitHub App that:

1. Hooks into your repositories and watches for new commits and releases
2. Analyzes commit messages using modern AI models
3. Categorizes changes based on what actually matters to your users
4. Generates clean changelogs in multiple formats for different audiences
5. Commits back to your repo or posts to your website automatically

The surprising discovery? When AI analyzes good commit messages, it can generate sensible categorizations that are actually useful to end users. That's not perfect — but it's dramatically better than the exhausted Friday afternoon "I'll just copy-paste from Slack" approach.

The Technical Approach

Under the hood, KittyLog is built on a modern developer stack:

1. GitHub Webhooks trigger on pushes/releases
2. FastAPI/Next.js backend fetches commit data via GitHub API
3. AI models analyze commit messages and classify changes
4. Templates format output for different audiences (developers vs end users)
5. Git automation commits generated changelogs back to repositories

The "magic" is in the prompt engineering — teaching the AI to distinguish between internal technical changes vs what actually matters to users.

Early Days, Honest Expectations

Right now, KittyLog is in early access with a small group of users. We're focused on the core functionality and actively improving:

• AI prompt tuning to better understand different types of projects
• Improved webhooks and real-time updates
• Reliability and error handling for edge cases
• Documentation and onboarding experience

The AI occasionally misinterprets what a commit "actually means" - that's where the prompt tuning work comes in. But for the core use case — turning messy git history into something readable for users — it's already saving users hours every month.

Try It Yourself (No Credit Card Required)

If you're tired of writing changelogs by hand too, you can try KittyLog for free:

• Free tier: 1 repository, 100 AI calls per month
• GitHub App installation: Takes literally 5 minutes
• No credit card required: Seriously, just install and try it

We're looking for early users to help shape the product - your feedback directly influences what we build next.

What's Next?

My roadmap is driven by actual user needs right now, not venture capital pitch decks. If you try it and hate it, tell me why. If you love it but need one specific feature, that's probably what I'll build next.

Sometimes the best tools come from solving your own problems. This started as a "let me automate this annoying task" project and is slowly growing into something other developers might find useful too.

---

If you're a developer who maintains projects and writes changelogs, I'd love to hear your workflow challenges. Drop a comment or reach out - the more I understand how different teams handle this, the better KittyLog can become.