Natural Language Prompts

The power of Billx-Agent lies in its ability to turn everyday language into valid, secure SQL queries. To get the most accurate and useful results, follow these guidelines when crafting your prompts.


✅ DO

✔️ Be Clear and Specific

Use straightforward, targeted prompts:

"Show me sales for Q1 2024 by product"

This helps the LLM accurately identify the intent and generate the correct SQL.


✔️ Use Common Data Terms

Use language that maps clearly to database fields:

date, total, amount, quantity, status, region, etc.

"What is the total amount of orders by region?"


✔️ Add Filters Where Needed

Use natural filter expressions to scope your data:

"List orders over $500 from March"

"Show users who signed up in the last 30 days"


✔️ Use Defined Timeframes

The AI understands relative timeframes such as:

  • "last 7 days"

  • "current month"

  • "past year"

  • "first quarter of 2023"


❌ AVOID

❌ Vague or Open-Ended Prompts

Avoid prompts that lack a specific objective:

"Give me something interesting" "Analyze the data" "What can you tell me?"

These don’t map clearly to SQL logic and may produce errors or generic output.


❌ Requests That Imply Write Access

Billx-Agent only supports read-only SQL (SELECT queries). Do not request modifications:

"Delete old orders" "Update all user emails" "Insert a new record"

Such queries will be blocked for safety.


❌ Prompt Overload

Avoid stacking too many conditions or objectives into one prompt:

"Show sales, profit, inventory, and shipment delays for all products in 3 regions, excluding weekends, and grouped by supplier and month"

🧠 Tip: Break large queries into multiple, focused prompts.


✅ A well-phrased prompt = faster results, cleaner SQL, and lower token usage.

Last updated