Cost controls
Keep Doram's load on your warehouse within a budget.
Background jobs
Background jobs keep the graph current. They read metadata and log tables, profile columns from samples instead of full scans, and replay approved metrics on a recent window.
You set a monthly compute budget and a time window for these jobs. You can also give Doram its own warehouse or resource group. For a production database, connect Doram to a read replica, never the primary, so its jobs and questions can't slow down production.
Jobs run in order of priority. If the budget runs out, the least important jobs pause first, and the replay of approved metrics stops last.
Each question
A question costs one bounded read of the graph, one LLM call to read the question, one warehouse query, and the LLM tokens used to explain the answer. Before the query runs, the compiler gets a cost estimate from your warehouse. If the query would cost too much, Doram asks for a narrower question, such as a shorter time range.