author here! Good suggestion, we should probably come up with some GOAL.md examples. With that said, one-shotting a pretty large app is a somewhat doable task, and that's one of the reasons we have introduced the interview step: exactly to let the model pull from you (instead you pushing into the model a spec document) what it needs to know to be able to work autonomously.
author here! It supports multi-repository. You would need to create a directory with both git repositories cloned in, and save the GOAL.md at the parent. This UX could use some polish, for sure. It works, but it needs this extra step.
author here! I am not sure if I would phrase it as sane Gas Town. The design constraints are very different. For example, I tried to optimized for output density (the number of tokens that I didn't have to rewrite because the output quality was bad), instead of treating it as a distributed system and coordination problem. SGAI, for example, runs one agent at a time - maybe one day it will run more agents in parallel, but only as much that I can obtain better outputs (not more outputs).
Also, I am not a cryptocurrency enthusiast - and therefore, I wouldn't accept creating a memecoin out of it.
author here! it is my daily driver for quite some time; with that said, its current shape is a bit of a more recent development. Initially, I would manually handle jj workspaces and fire out screen/tmux sessions; but over time, I figure it would be nice to have an UI that I could browse from anywhere through a VPN.
It does take some investment -- by adding customizations through the overlay folder (`sgai/` directory at the root of the repository) -- but eventually it should be able to code in a way that you would approve in a PR.
author here! the decision was mine; if anything, the senior leadership was fine with an unencumbered open-source license. What I didn't want was someone using it to make a business out of this tool without me in the mix.
In a sense, a futile effort; because if you reverse engineer a nlspec and rebuild it, then you can have it with any license you may want.
I have been working on this for my own use until recently, when I shared with the rest of the team, and we thought it would be nice to let the world see it.
I have been interested in autonomous code development for quite some time (at least since March/April 2025) - and summer '25 is when I felt the models were good enough to be pushed to autonomy.
I wrote a bit about it[0], and sgai is the incarnation of my take on AI autonomous coding.
sgai is not even v0 yet, a lot of work to be done to improve its implementation - but I think it should be usable enough for those willing to give it a try.