GAAI — Governed Agentic AI Infrastructure
A folder of Markdown, YAML and bash you drop into a git project to add cross-session memory and delivery governance to AI coding tools, splitting a discovery step that plans from a delivery step that executes in isolation.
GAAI adds a .gaai folder to a project with no SDK, package or external service required, vendoring its own Python runtime and a few Node helpers. A Discovery agent works with the developer in the current session to clarify what to build and writes a Story with acceptance criteria to a backlog; Discovery only reasons and never executes.
A separate Delivery agent executes a Story, always in an isolated OS process launched through tmux, so its system prompt cannot be contaminated by the Discovery session; it never decides scope on its own. A daemon can poll the backlog and deliver multiple stories in parallel, each in its own tmux session, addressing the problem of an AI coding agent touching code it should not, forgetting prior decisions, or shipping work no one can verify against criteria.
It works with Claude Code, Codex CLI, Gemini CLI, Cursor and other agents.