An open-source project, building in public

AI agents changed how fast software gets written. What did not change is who is accountable for what ships. Most teams supervising agent work today do it with scrollback and good intentions: no persisted plan, no enforced limits, no independent check before merge.

We think that gap needs a control system, the way money movement needed one. So Keelson runs every agent change through one supervised loop: a request becomes a plan a person approves, execution is constrained server-side (branch-only writes, scoped paths, no secrets, no merge or deploy authority), an independent review grades the result, and a person decides what ships. Underneath sit real Git hosting, an encrypted Vault, and an interface with both a plain-language and an engineer’s view, because supervision only works when everyone accountable can actually read the controls.

We win when a small team can honestly say where every agent-made change came from, who approved it, and why it shipped, and when the engineer on that team never feels the platform is a toy.