Natural Language Queries

Natural Language Queries

The core capability of AI for Database is translating plain-English questions into database queries. This guide covers how it works, best practices, and dozens of real examples.

How Translation Works

AI for Database reads your database schema (table names, column names, data types, foreign keys, and indexes) and sends this context along with your question to a large language model. The model generates a SQL query (or MongoDB aggregation pipeline) that answers your question.

The schema context is refreshed automatically when tables change. You can also trigger a manual refresh from Settings > Connections > Refresh Schema.

Example Questions by Category

Counting and Aggregation

• "How many orders were placed today?"
• "What is the total revenue for Q1 2024?"
• "What is the average response time for support tickets this week?"

Filtering and Searching

• "Show me all customers in California with more than 5 orders"
• "Find invoices that are overdue by more than 30 days"
• "Which products have less than 10 units in stock?"

Grouping and Ranking

• "Top 10 customers by lifetime value"
• "Revenue by product category for each month in 2024"
• "Number of new signups by country, ordered from highest to lowest"

Comparisons and Trends

• "Compare this month's sales to last month's"
• "Show the trend of daily active users over the past 90 days"
• "Which region grew the fastest in revenue this quarter?"

Joins and Relationships

• "List each customer's name and their most recent order date"
• "Show me the product name and category for all items with low stock"
• "How many tickets has each support agent closed this month?"

Handling Ambiguity

When AI for Database encounters ambiguity, it makes a best guess and shows its reasoning. For example, if you ask "How many users?" and your database has both a users table and a customers table, the AI will pick the most likely one and explain its choice.

You can resolve ambiguity by being explicit: "How many rows in the customers table?" or by setting table aliases in Settings to map business terms to specific tables.

Follow-Up Questions

After getting an initial result, you can refine with follow-up questions in the same conversation:

1. 1"Show me total sales by region" (initial query)
2. 2"Now filter to only regions with sales above $100K" (follow-up)
3. 3"Sort by sales descending" (another follow-up)

Each follow-up builds on the previous context, making it easy to explore your data iteratively.

Viewing and Editing SQL

Click View SQL on any result to see the generated query. You can edit the SQL directly and re-run it. This is useful for fine-tuning results or learning SQL by example.

Edited queries can be saved and will preserve your modifications on future runs.