How to Connect LangGraph to a SQL Database for Data-Driven Workflows
Learn how to build a custom SQL agent with LangGraph to connect to SQL databases, answer data questions, and ensure safety with human-in-the-loop...
Direct answer
To connect LangGraph to a SQL database for data-driven workflows, build a custom SQL agent via steps like installing libraries, selecting a tool-capable LLM, configuring the database, adding interaction tools, defining workflow steps with LangGraph’s Graph API, and implementing the agent. Key safety measures include limiting database permissions and using LangGraph’s human-in-the-loop feature to review queries before execution.
LangGraph, a framework for building stateful, multi-step AI applications, enables the creation of custom SQL agents to interact with SQL databases and answer data-related questions. These agents streamline data-driven workflows by automating querying and analysis tasks.
Building a LangGraph SQL agent involves core steps: install necessary libraries for LangGraph and database interaction, select an LLM that supports tool calls (like OpenAI’s gpt-5.5 or Anthropic’s claude-sonnet-4-6), configure the database (e.g., SQLite with the chinook sample database), add tools such as sql_db_list_tables, sql_db_schema, and sql_db_query, define the workflow using LangGraph’s Graph API (state, nodes, edges, conditional edges), and implement the agent.
Customization is key—you can create dedicated nodes for specific tool calls and tailor prompts for each step. For safety, limit database permissions to essential operations and use LangGraph’s human-in-the-loop feature to pause and get human approval before executing SQL queries.
Compatible LLMs include OpenAI’s gpt-5.5, Anthropic’s claude-sonnet-4-6, Azure’s gpt-5.5, Google Gemini’s gemini-2.5-flash-lite, AWS Bedrock’s us.anthropic.claude-sonnet-4-6, HuggingFace’s microsoft/Phi-3-mini-4k-instruct, and OpenRouter’s auto.
Sources
- LangChain Official Documentation: https://docs.langchain.com/
- LangGraph SQL Agent Example Code: https://github.com/langchain-ai/docs/edit/main/src/oss/langgraph/sql-agent.mdx
- LangGraph Overview Documentation: https://docs.langchain.com/oss/python/langgraph/overview
- SQLite Sample Database: https://www.sqlitetutorial.net/sqlite-sample-database/
FAQ
- What tools are used to interact with SQL databases in LangGraph?
- Tools like sql_db_list_tables (to list tables), sql_db_schema (to retrieve table schemas), and sql_db_query (to run queries) are commonly used for SQL database interactions in LangGraph.
- How can I ensure SQL database safety when using LangGraph?
- Minimize database connection permissions to essential operations and use LangGraph’s human-in-the-loop feature to pause and get human approval before executing SQL queries.
- Which LLMs are compatible with LangGraph for SQL agents?
- Compatible LLMs include OpenAI’s gpt-5.5, Anthropic’s claude-sonnet-4-6, Azure’s gpt-5.5, Google Gemini’s gemini-2.5-flash-lite, AWS Bedrock’s us.anthropic.claude-sonnet-4-6, HuggingFace’s microsoft/Phi-3-mini-4k-instruct, and OpenRouter’s auto.
- What is the LangGraph Graph API used for?
- The LangGraph Graph API helps define workflows using concepts like state, nodes, edges, and conditional edges, which are essential for building custom SQL agents.