Draxlr in AI Database Agents

Draxlr made the questions visible and attributable to the people who asked them. The admin could see all five test chats in the sidebar and identify which account asked each question.

Draxlr gave a single-value answer in plain language. It stated that there were 308 orders in total.

Draxlr handled the follow-up correction by using the updated month instead of the original one. The second reply said June 2026, and the second query was limited to June only; it did not combine July with June or keep July in place.

Draxlr did not ask for clarification when the follow-up was ambiguous. After the first reply produced two plausible customer sets, the second reply answered directly and the transcript shows no clarifying question, option prompt, or disambiguation before it.

Draxlr kept the previous answer’s customer set intact in the follow-up. It answered for the same five customers, did not ask who “they” were, and the second query rebuilt that same set before counting their last-90-day orders.

Draxlr fails this scenario: when the needed table is excluded, it still answers from another table instead of saying it cannot answer. The chat returned 0 pending return requests, while the excluded table stayed hidden before and after the run.

Draxlr passed this request to gather several findings in one place. It gave the three direct answers, then combined them into one report without changing the values; the saved report reopened with the same single row.

Draxlr generalized a trusted question-and-query example to a rephrased question. After training, both phrasings returned 42, the trusted count, and the rephrased question used the trusted query’s logic.