Open Source · Multi-Agent AI
Natural language questions. Instant SQL. No SQL knowledge required.
NL-2-SQL Agents is an open source AI agent system that translates plain-English questions into accurate SQL queries, refines them autonomously, and explains every step — across any major database.
What It Is
A Multi-Agent Pipeline for Accurate SQL Generation
NL-2-SQL Agents orchestrates specialized AI agents that collaborate to turn any natural language question into a correct, optimized SQL query — with full explainability.
Schema Discovery Agent
Automatically introspects your database schema, understands table relationships, and builds a semantic context layer for accurate query generation.
Query Generation Agent
Translates the natural language question into SQL using the schema context, with support for complex JOINs, aggregations, and subqueries.
Refinement Agent
Validates and iteratively improves the generated SQL against the actual schema, catching errors before execution.
Explanation Agent
Generates a plain-English explanation of what the SQL query does — making every result transparent and auditable.
Capabilities
Everything You Need for Natural Language Data Access
NL-2-SQL Agents is built for developers who want to embed intelligent SQL generation into their applications.
Multi-Database Support
Works with PostgreSQL, MySQL, SQLite, Snowflake, and BigQuery out of the box. Pluggable connectors for additional databases.
Agentic Query Refinement
Agents autonomously retry and refine queries that fail validation, dramatically improving accuracy on complex schemas.
Schema-Aware Context
Automatically builds a semantic understanding of your schema — table names, column types, foreign keys, and relationships.
Explainable Results
Every generated query comes with a plain-English explanation. Show your users exactly what data is being retrieved and why.
Embeddable API
Expose NL-2-SQL as a REST API or integrate it directly into your Java, Python, or Node.js application with a clean SDK.
Audit & Safety
Read-only query enforcement, query approval workflows, and full audit logging — safe to deploy in production environments.
LLM Agnostic
Works with OpenAI, Anthropic, Azure OpenAI, and local LLMs via Ollama. Swap models without changing your application code.
MIT Licensed
Fully open source. Self-host it, embed it, extend it — free to use with attribution.
How It Works
NL-2-SQL Agents follows a structured multi-agent pipeline for every query.
Step 1: Connect your database
Provide your database connection string. The Schema Discovery Agent introspects your tables, columns, types, and relationships automatically.
Step 2: Ask in plain English
A user types a question like "Show me total revenue by region for Q1 2025." No SQL knowledge needed.
Step 3: Agents generate & refine
The Query Generation Agent produces the SQL. The Refinement Agent validates it against the schema and iterates until correct.
Step 4: Execute & explain
The query runs safely (read-only by default). The Explanation Agent returns results alongside a plain-English breakdown of what was queried.
Integration
Embed in Minutes
NL-2-SQL Agents is designed to be embedded directly into your existing applications.
# Python example from nl2sql_agents import NL2SQLPipeline pipeline = NL2SQLPipeline( db_url="postgresql://...", llm="openai/gpt-4o" ) result = pipeline.query( "What were our top 10 customers by revenue last quarter?" ) print(result.sql) # Generated SQL print(result.data) # Query results print(result.explanation) # Plain-English explanation
REST API
Deploy as a standalone REST service and call it from any language or framework.
Python & Java SDKs
Native client libraries for Python and Java with full async support.
Streaming Results
Stream query results and explanations in real time for responsive UIs.
Open source. Self-hostable. Free.
MIT licensed and free to embed in your applications. Star us on GitHub and help shape the roadmap.