Project structure

After cloning the private repository (Pro buyers only), the layout is:

ai-cost-gate/
├── cmd/
│   └── aicg/                  # main binary entry point
├── internal/
│   ├── proxy/                 # OpenAI-compatible proxy
│   ├── attribution/           # project/agent/model attribution
│   ├── budget/                # budget, throttle, auto-stop
│   ├── store/                 # SQLite persistence
│   ├── admin/                 # admin HTTP API
│   └── dashboard/             # embedded web dashboard
├── web/                       # dashboard frontend (assets)
├── data/                      # SQLite file lives here at runtime
├── docker-compose.yml
├── Dockerfile
├── .env.example
└── README.md

Top-level conventions

• cmd/ holds the entry points. There is only one binary, aicg, which starts the proxy, the admin API, and the embedded dashboard.
• internal/ holds all business logic. Nothing in internal/ is importable from outside this module — that's Go's convention and we use it.
• web/ contains the static assets for the dashboard. They are embedded into the binary with go:embed so the docker image stays as a single artifact.

Where to look

• Adding a provider adapter? Start at internal/proxy/adapters/.
• Changing how cost is calculated? Look at internal/attribution/pricing.go.
• Changing the auto-stop heuristic? Look at internal/budget/autostop.go.
• Changing the schema? Look at internal/store/migrations/.

Build artifacts

After go build ./cmd/aicg, the binary is at ./aicg. The Docker image is built with a multi-stage Dockerfile that uses a distroless base — no shell, no package manager, ~30MB final image.