Turn off agent & assistant auto-commit

I don’t like how my git branch is flooded with commits for every single action the agent and assistant take, even my rollbacks of those actions. I’d like the option to only choose to commit changes when I’m happy with them and keep my git branch nice and clean.

5 Likes

I won’t say I feel the same way, but I will say I have similar thoughts about the checkpoints and billing methods. But, I also understand the costs of LLM, and win lose or draw, it costs money to operate these things, especially with drastically long context.

I think the REAL problem is context window being too small, so you feel like you’re looping because, well, you are. But it’s a limitation of our time, and I don’t believe malicious or intentional. It’s just the current iteration and implementation isn’t perfect.

Hopefully we progress quickly past this point, because we’re currently in a frustrating period.

1 Like

I should clarify. I understand the business model they have, and that they charge $0.25 per Agent checkpoint and $0.05 per Assistant checkpoint. Would it be great if it was even cheaper? Sure. But that’s not my feedback. I also don’t believe that it’s intentionally malicious.

My feedback isn’t in the cost structure, but in the git commit graph management. My ideal would be to still have the “checkpoints” for easy rollbacks, but to not have changes committed to the branch until I approve them. Having a branch of 1,000+ commits including rollbacks is just difficult to manage.

At least let me name the commit, or give me the option to, through the agent/assistant checkpoint interface.

1 Like

Omg yes it’s useless to have git hooked up when every single update is checked in. It wasn’t like this before it’s only when the update to the new agent came. I used to be able to name commits and bundle them into logical things.

In general the inability to start work on an idea and roll back and easily understand what files were changed is a problem. I find myself deep into a feature but with a bunch of iteration and work left to do when I have a bug I want to fix but can’t since the whole code isn’t ready to deploy so I either have to wait to finish my bigger initiative or abandon it and revert files back a few days for anything I think I’ve touched. The rollbacks are helpful but they don’t save that many.

Hanging the ability to actually reset to a git commit would be interesting too.

1 Like