I wanted to share with you all my experience on coding with AI, on a self-hosted oss.
This entire post has been fully written by me without any AI.
Edit: I violated rule 4, so I cut the long text here, please read it on the linked blog post!
Also, to be clear, the post itself has no ads and no monetization whatsoever. This is not self promotion for any scope than to share my experience on a controversial topic.
Italian is a beautiful language, probably my favorite one.
As far as your post is concerned, I’m using qwen 3.8 27B as my main local LLM. I actually do have it push code to a github repo. Mainly because I have no real interest in learning the code it is writing, I’ve always been a back end and data person and I’m having it write web stuff. One thing I do a little differently is I make sure the model writes documentation up front and then at the end and also any steps it discovers, that it should remember into an MD file. This is so it doesn’t make the same mistakes over and over.
I also make sure it writes instructions for itself(or another model) and I will sometimes have another model check its work.
I have started implementing long term memory with obsidian as its “second brain”. This way I don’t have to keep explaining who I am or what we’re doing, what machines are available, what my preferences are, etc…
I think in a few weeks I’ll be able to grab my phone and start giving it tasks to do, using Hermes.
Nice write up. I appreciate you taking the time to share.
Can you elaborate on using obsidian as a memory store?
Try this out https://www.youtube.com/watch?v=4Hvkv_I8QDE
This is where I learned of it, but I’ve also hooked up unsloth desktop to the same obsidian knowledgebase. This way the long term memory persists across my LLM client tools. Right now I have it keep track of me and my preferences, my wife and her preferences, home lab architecture and documentation and technical project information for me and my wife.