← Docs
Recipe

Natural Language Database Query

Ask questions in plain English and get structured results from your database — no SQL required.

Overview

This recipe wires a natural language input to a schema-aware LLM pipeline that generates, validates, and executes SQL against your live database. Results stream back as formatted tables.

Ingredients

  • Schema introspection endpoint (tables, columns, types, foreign keys)
  • LLM prompt template with schema context injection
  • SQL validator and read-only guard
  • Result formatter with column type awareness

Steps

  1. 1Introspect schema. Fetch table metadata and serialize it as a compact DDL-like summary for the prompt.
  2. 2Build the prompt. Combine the user question, schema summary, and strict output-format instructions into a single LLM call.
  3. 3Validate the SQL. Reject any statement that is not a SELECT. Check table and column existence against the schema.
  4. 4Execute and format. Run the query with a row limit and timeout. Render results as a responsive table with proper type formatting.

Security note: Always enforce read-only access. Never allow INSERT, UPDATE, DELETE, DROP, or ALTER. Use a dedicated database user with SELECT-only privileges.

Back to DocsAll Recipes