Every client gets an AI spend statement, per account

Your agency runs an agent per client, each in that client's own container on one flat-rate box. The server costs the same whether twelve agents run or forty.

The tokens don't. Each client's agent calls the Hoody AI Gateway from its own container, 5% over provider cost, drawn from your AI Balance. At month-end you GET /api/v1/agent/statistics against each client's container: totalCostUsd, with the by-model split underneath. Client seven's agent burned $41 in tokens; client nineteen's burned $4. Those numbers go on each invoice as a pass-through line, with the statistics response as backup.

Before, AI spend was one opaque OpenAI bill you split by guesswork. Now the attribution is structural: one client, one container, one statistics call.