Project Study Case
The Problem
The Proposal
The Architecture
Solution Highlights
Big Commits? We got you.
This technique can:
- Make Code Reviews much easier to do
- Enable bisecting to work properly, making debugging much easier
- Make rollbacks less intensive preventing cherry-picks
- Allow to easily identify the intent of each change, making the git history more readable and informative
Best Practices everywhere
Good Prompt = Cheaper Models
Results
No More Messy Histories
- Vague or inconsistent commit messages are reduced to 0 in projects.
- Things done that can be forgotten between a commit and another doesn't exists anymore.
- Even large, multi-purpose commits are automatically split into atomic, logically-grouped commits.
Consistent Best Practices
- Multiple practices in project (gimoji, semantic, numbered) are reduced to a single pattern
- Blaming and bisecting work properly, making debugging much easier
- Reverting changes is easier and less error-prone.
No time waste by waiting
- Loading the prompt is faster than writing the commit message.
- The loading for the whole process is around 2-3 seconds, in large commits.
- By using the CLI, developers stay in flow and don't have to context-switch to write commit messages
Code Quality and Team Collaboration
- Code reviews are more effective with clear, atomic commits.
- Teams have a more consistent and informative git history, improving the onboarding of new members
- Overall code quality is improved as the commit history becomes a reliable source of information about the codebase's evolution.
Future Improvements
Reduce the Bundle Size
- Rust ships the same environment with ~2MB executable
- Reduces bundle-size by 97%
- Easier onboarding with
aptorhomebrew - Rust is easier to maintain, but has a bigger learning curve.
Local-First AI integration
- Kimi, Minimax, Qwen and GLM can read code easily, providing even more privacy.
- Responses without latency, pure TPS.
- No trolls = Pure internal processing.
- No skills or black magic for token saving.