Natural Language to SQL: Querying Enterprise Databases with AI

Natural-language-to-SQL translates plain English questions into read-only SQL queries against your databases, using schema metadata to ground the translation and a validation layer to block destructive statements. ZenSearch supports PostgreSQL, MySQL, ClickHouse, and SQL Server; failed queries are auto-corrected by feeding the error back to the LLM with the relevant schema so most mistakes are resolved in a single retry.

Not all enterprise knowledge lives in documents. Critical data sits in relational databases — sales figures, customer records, inventory levels, operational metrics. ZenSearch's NL-to-SQL capability lets users query these databases by asking questions in plain English.

How It Works

1. Schema Discovery — When you connect a database, ZenSearch introspects the schema: tables, columns, data types, and relationships. This metadata helps the AI understand your data model and generate accurate queries.

2. Query Translation — When a user asks "What were our top 10 customers by revenue last quarter?", the AI uses the schema context to generate the appropriate SQL query.

3. Safety Validation — Before execution, every query is validated against safety rules:

• No destructive operations (DROP, DELETE, INSERT, UPDATE)
• No schema modifications (CREATE, ALTER)
• Query execution timeout limits
• Row count limits to prevent runaway queries

4. Execution — The validated query runs in read-only mode against your database. Results are returned as structured data.

5. Synthesis — The AI formats the results into a natural language response with the data presented clearly — tables, summaries, or narrative form depending on the question.

Error Correction

When a generated SQL query fails — wrong column name, invalid join, syntax error — the error is automatically fed back to the AI with schema context. The query is corrected and retried. Most errors are resolved in a single correction cycle.

Supported Databases

• PostgreSQL — Full support including SSL/TLS and schema filtering
• MySQL — TLS support, charset configuration, table and view filtering
• ClickHouse — Optimized for analytics workloads and large-scale data
• SQL Server — Windows and SQL authentication, encrypted connections

Combined with Document Search

The real power emerges when database queries combine with document search in a single conversation. For example, you can ask: "Which customers had the most support tickets last quarter, and what does our escalation policy say about high-volume accounts?" The AI agent searches your ticket database for the data, then searches your documentation for the policy — delivering a complete, sourced answer.