What do you want to know?

Ask a business question about 100,000 real orders. Plainsight writes the SQL, checks it can only read, runs it and shows you the answer with the query behind it.

One question in

“What percentage of delivered orders were late?”

SQL written

SELECT avg(order_delivered_customer_date
         > order_estimated_delivery_date)
FROM orders
WHERE order_status = 'delivered'

Answer out

8.1% of 96,470 orders

20 questions left this hour · answered by gpt-5.4-mini

Tested on 60 questions with checked answers: 85% right with table names alone, 100% once the model is given written notes about the data.

How an answer is made

  1. The model writes one query

    It gets your question, written notes about the tables and their traps, and six worked examples. If the data can't answer the question, it says so instead of guessing.

  2. Code checks it

    The query is parsed, and anything other than a single SELECT is refused before it reaches the database.

  3. It runs read-only

    The database is opened read-only, can't read files or web addresses, and stops after 15 seconds or 1,000 rows.

  4. One second try

    If the query fails, the error goes back to the model once for a fix. You always see the SQL that produced the answer.