I've found it helps PR reviewers when they can look through a set of commits with clear messages and logically organized changes. Typically reviewers prefer a larger quantity of smaller changes versus a smaller quantity of larger changes. Sometimes it gets really messy to break up a change into sufficiently small PRs, so thoughtful commits are a great way of further subdividing changes in PRs. It can be pretty time consuming to do this though, so this tool automates the process with the help of AI.

The tool sends the diff of your git branch against a base branch to an LLM provider. The LLM provider responds with a set of suggested commits with sensible commit messages, change groupings, and descriptions. When you explicitly accept the proposed changes, the tool re-writes the commit history on your branch to match the LLM's suggestion. Then you can force push your branch to your remote to make it match.

The default AI provider is your locally running Ollama server. Cloud providers can be explicitly configured via CLI argument or in a config file, but keeping local models as the default helps to protect against unintentional data sharing. The tool always creates a backup branch in case you need to easily revert in case of changing your mind or an error in commit re-writing. Note that re-writing commit history to a remote branch requires a force push, which is something your team/org will need to be ok with. As long as you are working on a feature branch this is usually fine, but it's always worth checking if you are not sure.