Warning
This is clearly a work in progress thing. This whole project is an excuse for me to learn a bit more about self hosted LLM and integration in local apps since this subject is directly connected to my enterprise job
Globally, just do your commit message yourself, if you have work on it, you should be able to resume it in a commit message ;)
Goomit is an auto commit generator that uses LLM to generate commit message depending on your git diff command result.
This is more a POC than anything else but if it works that's cool !
Download the model you want using ollama ollama pull llama3
Then just use the app, you can specify the model by using the -m argument, default one is llama3
You can also generate a sort of config that summarize some information about your repo (readme, used language, repo description) so the prompt will include this context for better results
- go mod tidy
- ./build.sh
- ./goomit config generate
- ./goomit -m my-model
Since this is a work in progress there is some limitations :
- Prompt isn't very polished yet and commit message can be inconsistent, too much verbose, paraphrasing the code line by line
- I need to run some test by using different model so I can use the one I feel like is the best as the default one
- The model as no context at all about your project and this is a feature I need to implement so the commit message is more accurate
- Some bugs may or may not occurs, as I said this is not a production ready things but just a POC
Since the last propmpt update, a new important limitation has appeard. There is huge performance issue depending on the model you'r using and your machine capability. It's not a weird behavior when using self hosted llm but still this is a huge jump in time needed for commit generation between old and new prompt. Will need to investigate on that.
This project is entirely made by me (ASTOLFI Vincent). I suggest you to check on my github profile if you want to see the other project I've done for my studies or the ones I do in my free time.