We periodically collect information about inference tokens used by customers. This is done by periodically sending a request to inference gateway, which responds with tokens used since the last data exchange, at per project granularity. Token consumption data collected that way are then turned into `InferenceTokens` billing records.