构建 SQL 代理

概述

在本教程中，你将学习如何构建一个能够使用 LangChain 代理回答关于 SQL 数据库问题的代理。从高层次来看，该代理将：

从数据库中获取可用的表和模式

决定哪些表与问题相关

获取相关表的模式

根据问题和模式信息生成查询

使用 LLM 双重检查查询中的常见错误

执行查询并返回结果

纠正数据库引擎发现的错误，直到查询成功

根据结果制定响应

构建 SQL 数据库的问答系统需要执行模型生成的 SQL 查询。这样做存在固有风险。请确保你的数据库连接权限始终尽可能地限定在代理所需的最小范围内。这将减轻（但不能消除）构建模型驱动系统的风险。

概念

我们将涵盖以下概念：

用于从 SQL 数据库读取的工具
LangChain 代理
Human in the Loop 流程

设置

安装

pip install langchain  langgraph  langchain-community

LangSmith

设置 LangSmith 以检查你的链或代理内部发生的情况。然后设置以下环境变量：

export LANGSMITH_TRACING="true"
export LANGSMITH_API_KEY="..."

1. 选择一个 LLM

选择一个支持工具调用的模型：

👉 阅读 OpenAI 聊天模型集成文档

pip install -U "langchain[openai]"

import os
from langchain.chat_models import init_chat_model

os.environ["OPENAI_API_KEY"] = "sk-..."

model = init_chat_model("gpt-5.4")

import os
from langchain_openai import ChatOpenAI

os.environ["OPENAI_API_KEY"] = "sk-..."

model = ChatOpenAI(model="gpt-5.4")

👉 阅读 Anthropic 聊天模型集成文档

pip install -U "langchain[anthropic]"

import os
from langchain.chat_models import init_chat_model

os.environ["ANTHROPIC_API_KEY"] = "sk-..."

model = init_chat_model("claude-sonnet-4-6")

import os
from langchain_anthropic import ChatAnthropic

os.environ["ANTHROPIC_API_KEY"] = "sk-..."

model = ChatAnthropic(model="claude-sonnet-4-6")

👉 阅读 Azure 聊天模型集成文档

pip install -U "langchain[openai]"

import os
from langchain.chat_models import init_chat_model

os.environ["AZURE_OPENAI_API_KEY"] = "..."
os.environ["AZURE_OPENAI_ENDPOINT"] = "..."
os.environ["OPENAI_API_VERSION"] = "2025-03-01-preview"

model = init_chat_model(
    "azure_openai:gpt-5.4",
    azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"],
)

import os
from langchain_openai import AzureChatOpenAI

os.environ["AZURE_OPENAI_API_KEY"] = "..."
os.environ["AZURE_OPENAI_ENDPOINT"] = "..."
os.environ["OPENAI_API_VERSION"] = "2025-03-01-preview"

model = AzureChatOpenAI(
    model="gpt-5.4",
    azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"]
)

👉 阅读 Google GenAI 聊天模型集成文档

pip install -U "langchain[google-genai]"

import os
from langchain.chat_models import init_chat_model

os.environ["GOOGLE_API_KEY"] = "..."

model = init_chat_model("google_genai:gemini-2.5-flash-lite")

import os
from langchain_google_genai import ChatGoogleGenerativeAI

os.environ["GOOGLE_API_KEY"] = "..."

model = ChatGoogleGenerativeAI(model="gemini-2.5-flash-lite")

👉 阅读 AWS Bedrock 聊天模型集成文档

pip install -U "langchain[aws]"

from langchain.chat_models import init_chat_model

# 按照此处步骤配置您的凭证：
# https://docs.aws.amazon.com/bedrock/latest/userguide/getting-started.html

model = init_chat_model(
    "anthropic.claude-3-5-sonnet-20240620-v1:0",
    model_provider="bedrock_converse",
)

from langchain_aws import ChatBedrock

model = ChatBedrock(model="anthropic.claude-3-5-sonnet-20240620-v1:0")

👉 阅读 HuggingFace 聊天模型集成文档

pip install -U "langchain[huggingface]"

import os
from langchain.chat_models import init_chat_model

os.environ["HUGGINGFACEHUB_API_TOKEN"] = "hf_..."

model = init_chat_model(
    "microsoft/Phi-3-mini-4k-instruct",
    model_provider="huggingface",
    temperature=0.7,
    max_tokens=1024,
)

import os
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint

os.environ["HUGGINGFACEHUB_API_TOKEN"] = "hf_..."

llm = HuggingFaceEndpoint(
    repo_id="microsoft/Phi-3-mini-4k-instruct",
    temperature=0.7,
    max_length=1024,
)
model = ChatHuggingFace(llm=llm)

👉 阅读 OpenRouter 聊天模型集成文档

pip install -U "langchain-openrouter"

import os
from langchain.chat_models import init_chat_model

os.environ["OPENROUTER_API_KEY"] = "sk-..."

model = init_chat_model(
    "auto",
    model_provider="openrouter",
)

import os
from langchain_openrouter import ChatOpenRouter

os.environ["OPENROUTER_API_KEY"] = "sk-..."

model = ChatOpenRouter(model="auto")

下面示例中显示的输出使用了 OpenAI。

2. 配置数据库

你将为本教程创建一个 SQLite 数据库。SQLite 是一个轻量级数据库，易于设置和使用。我们将加载 chinook 数据库，这是一个代表数字媒体商店的示例数据库。为方便起见，我们已将数据库 (Chinook.db) 托管在公共 GCS 存储桶中。

import requests, pathlib

url = "https://storage.googleapis.com/benchmarks-artifacts/chinook/Chinook.db"
local_path = pathlib.Path("Chinook.db")

if local_path.exists():
    print(f"{local_path} already exists, skipping download.")
else:
    response = requests.get(url)
    if response.status_code == 200:
        local_path.write_bytes(response.content)
        print(f"File downloaded and saved as {local_path}")
    else:
        print(f"Failed to download the file. Status code: {response.status_code}")

我们将使用 langchain_community 包中提供的便捷 SQL 数据库包装器与数据库交互。该包装器提供了一个简单的接口来执行 SQL 查询和获取结果：

from langchain_community.utilities import SQLDatabase

db = SQLDatabase.from_uri("sqlite:///Chinook.db")

print(f"Dialect: {db.dialect}")
print(f"Available tables: {db.get_usable_table_names()}")
print(f'Sample output: {db.run("SELECT * FROM Artist LIMIT 5;")}')

Dialect: sqlite
Available tables: ['Album', 'Artist', 'Customer', 'Employee', 'Genre', 'Invoice', 'InvoiceLine', 'MediaType', 'Playlist', 'PlaylistTrack', 'Track']
Sample output: [(1, 'AC/DC'), (2, 'Accept'), (3, 'Aerosmith'), (4, 'Alanis Morissette'), (5, 'Alice In Chains')]

3. 添加用于数据库交互的工具

使用 langchain_community 包中提供的 SQLDatabase 包装器与数据库交互。该包装器提供了一个简单的接口来执行 SQL 查询和获取结果：

from langchain_community.agent_toolkits import SQLDatabaseToolkit

toolkit = SQLDatabaseToolkit(db=db, llm=model)

tools = toolkit.get_tools()

for tool in tools:
    print(f"{tool.name}: {tool.description}\n")

sql_db_query: Input to this tool is a detailed and correct SQL query, output is a result from the database. If the query is not correct, an error message will be returned. If an error is returned, rewrite the query, check the query, and try again. If you encounter an issue with Unknown column 'xxxx' in 'field list', use sql_db_schema to query the correct table fields.

sql_db_schema: Input to this tool is a comma-separated list of tables, output is the schema and sample rows for those tables. Be sure that the tables actually exist by calling sql_db_list_tables first! Example Input: table1, table2, table3

sql_db_list_tables: Input is an empty string, output is a comma-separated list of tables in the database.

sql_db_query_checker: Use this tool to double check if your query is correct before executing it. Always use this tool before executing a query with sql_db_query!

4. 使用 `create_agent`

使用 create_agent 以最少的代码构建一个 ReAct 代理。该代理将解释请求并生成 SQL 命令，然后由工具执行。如果命令有错误，错误消息将返回给模型。然后模型可以检查原始请求和新的错误消息，并生成新的命令。这个过程可以持续进行，直到 LLM 成功生成命令或达到结束计数。这种向模型提供反馈（在这种情况下是错误消息）的模式非常强大。使用描述性的系统提示初始化代理以自定义其行为：

system_prompt = """
You are an agent designed to interact with a SQL database.
Given an input question, create a syntactically correct {dialect} query to run,
then look at the results of the query and return the answer. Unless the user
specifies a specific number of examples they wish to obtain, always limit your
query to at most {top_k} results.

You can order the results by a relevant column to return the most interesting
examples in the database. Never query for all the columns from a specific table,
only ask for the relevant columns given the question.

You MUST double check your query before executing it. If you get an error while
executing a query, rewrite the query and try again.

DO NOT make any DML statements (INSERT, UPDATE, DELETE, DROP etc.) to the
database.

To start you should ALWAYS look at the tables in the database to see what you
can query. Do NOT skip this step.

Then you should query the schema of the most relevant tables.
""".format(
    dialect=db.dialect,
    top_k=5,
)

现在，使用模型、工具和提示创建代理：

from langchain.agents import create_agent


agent = create_agent(
    model,
    tools,
    system_prompt=system_prompt,
)

5. 运行代理

在示例查询上运行代理并观察其行为：

question = "Which genre on average has the longest tracks?"

for step in agent.stream(
    {"messages": [{"role": "user", "content": question}]},
    stream_mode="values",
):
    step["messages"][-1].pretty_print()

================================ Human Message =================================

Which genre on average has the longest tracks?
================================== Ai Message ==================================
Tool Calls:
  sql_db_list_tables (call_BQsWg8P65apHc8BTJ1NPDvnM)
 Call ID: call_BQsWg8P65apHc8BTJ1NPDvnM
  Args:
================================= Tool Message =================================
Name: sql_db_list_tables

Album, Artist, Customer, Employee, Genre, Invoice, InvoiceLine, MediaType, Playlist, PlaylistTrack, Track
================================== Ai Message ==================================
Tool Calls:
  sql_db_schema (call_i89tjKECFSeERbuACYm4w0cU)
 Call ID: call_i89tjKECFSeERbuACYm4w0cU
  Args:
    table_names: Track, Genre
================================= Tool Message =================================
Name: sql_db_schema


CREATE TABLE "Genre" (
	"GenreId" INTEGER NOT NULL,
	"Name" NVARCHAR(120),
	PRIMARY KEY ("GenreId")
)

/*
3 rows from Genre table:
GenreId	Name
1	Rock
2	Jazz
3	Metal
*/


CREATE TABLE "Track" (
	"TrackId" INTEGER NOT NULL,
	"Name" NVARCHAR(200) NOT NULL,
	"AlbumId" INTEGER,
	"MediaTypeId" INTEGER NOT NULL,
	"GenreId" INTEGER,
	"Composer" NVARCHAR(220),
	"Milliseconds" INTEGER NOT NULL,
	"Bytes" INTEGER,
	"UnitPrice" NUMERIC(10, 2) NOT NULL,
	PRIMARY KEY ("TrackId"),
	FOREIGN KEY("MediaTypeId") REFERENCES "MediaType" ("MediaTypeId"),
	FOREIGN KEY("GenreId") REFERENCES "Genre" ("GenreId"),
	FOREIGN KEY("AlbumId") REFERENCES "Album" ("AlbumId")
)

/*
3 rows from Track table:
TrackId	Name	AlbumId	MediaTypeId	GenreId	Composer	Milliseconds	Bytes	UnitPrice
1	For Those About To Rock (We Salute You)	1	1	1	Angus Young, Malcolm Young, Brian Johnson	343719	11170334	0.99
2	Balls to the Wall	2	2	1	U. Dirkschneider, W. Hoffmann, H. Frank, P. Baltes, S. Kaufmann, G. Hoffmann	342562	5510424	0.99
3	Fast As a Shark	3	2	1	F. Baltes, S. Kaufman, U. Dirkscneider & W. Hoffman	230619	3990994	0.99
*/
================================== Ai Message ==================================
Tool Calls:
  sql_db_query_checker (call_G64yYm6R6UauiVPCXJZMA49b)
 Call ID: call_G64yYm6R6UauiVPCXJZMA49b
  Args:
    query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AverageLength FROM Track INNER JOIN Genre ON Track.GenreId = Genre.GenreId GROUP BY Genre.Name ORDER BY AverageLength DESC LIMIT 5;
================================= Tool Message =================================
Name: sql_db_query_checker

SELECT Genre.Name, AVG(Track.Milliseconds) AS AverageLength FROM Track INNER JOIN Genre ON Track.GenreId = Genre.GenreId GROUP BY Genre.Name ORDER BY AverageLength DESC LIMIT 5;
================================== Ai Message ==================================
Tool Calls:
  sql_db_query (call_AnO3SrhD0ODJBxh6dHMwvHwZ)
 Call ID: call_AnO3SrhD0ODJBxh6dHMwvHwZ
  Args:
    query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AverageLength FROM Track INNER JOIN Genre ON Track.GenreId = Genre.GenreId GROUP BY Genre.Name ORDER BY AverageLength DESC LIMIT 5;
================================= Tool Message =================================
Name: sql_db_query

[('Sci Fi & Fantasy', 2911783.0384615385), ('Science Fiction', 2625549.076923077), ('Drama', 2575283.78125), ('TV Shows', 2145041.0215053763), ('Comedy', 1585263.705882353)]
================================== Ai Message ==================================

On average, the genre with the longest tracks is "Sci Fi & Fantasy" with an average track length of approximately 2,911,783 milliseconds. This is followed by "Science Fiction," "Drama," "TV Shows," and "Comedy."

代理正确编写了查询，检查了查询，并运行了查询以提供其最终响应。

你可以在 LangSmith 跟踪中检查上述运行的所有方面，包括采取的步骤、调用的工具、LLM 看到的提示等。

（可选）使用 Studio

Studio 提供了一个“客户端”循环以及内存，因此你可以将其作为聊天界面运行并查询数据库。你可以提出诸如“告诉我数据库的模式”或“显示前 5 名客户的发票”之类的问题。你将看到生成的 SQL 命令和结果输出。有关如何启动的详细信息如下。

在 Studio 中运行你的代理

除了前面提到的包之外，你还需要：

pip install -U langgraph-cli[inmem]>=0.4.0

在你将运行的目录中，你需要一个包含以下内容的 langgraph.json 文件：

{
  "dependencies": ["."],
  "graphs": {
    "agent": "./sql_agent.py:agent",
    "graph": "./sql_agent_langgraph.py:graph"
  },
  "env": ".env"
}

创建一个文件 sql_agent.py 并插入以下内容：

#sql_agent.py for studio
import pathlib

from langchain.agents import create_agent
from langchain.chat_models import init_chat_model
from langchain_community.agent_toolkits import SQLDatabaseToolkit
from langchain_community.utilities import SQLDatabase
import requests


# Initialize an LLM
model = init_chat_model("gpt-5.4")

# Get the database, store it locally
url = "https://storage.googleapis.com/benchmarks-artifacts/chinook/Chinook.db"
local_path = pathlib.Path("Chinook.db")

if local_path.exists():
    print(f"{local_path} already exists, skipping download.")
else:
    response = requests.get(url)
    if response.status_code == 200:
        local_path.write_bytes(response.content)
        print(f"File downloaded and saved as {local_path}")
    else:
        print(f"Failed to download the file. Status code: {response.status_code}")

db = SQLDatabase.from_uri("sqlite:///Chinook.db")

# Create the tools
toolkit = SQLDatabaseToolkit(db=db, llm=model)

tools = toolkit.get_tools()

for tool in tools:
    print(f"{tool.name}: {tool.description}\n")

# Use create_agent
system_prompt = """
You are an agent designed to interact with a SQL database.
Given an input question, create a syntactically correct {dialect} query to run,
then look at the results of the query and return the answer. Unless the user
specifies a specific number of examples they wish to obtain, always limit your
query to at most {top_k} results.

You can order the results by a relevant column to return the most interesting
examples in the database. Never query for all the columns from a specific table,
only ask for the relevant columns given the question.

You MUST double check your query before executing it. If you get an error while
executing a query, rewrite the query and try again.

DO NOT make any DML statements (INSERT, UPDATE, DELETE, DROP etc.) to the
database.

To start you should ALWAYS look at the tables in the database to see what you
can query. Do NOT skip this step.

Then you should query the schema of the most relevant tables.
""".format(
    dialect=db.dialect,
    top_k=5,
)

agent = create_agent(
    model,
    tools,
    system_prompt=system_prompt,
)

6. 实现Human in the Loop审查

在执行代理的 SQL 查询之前进行检查，以防止任何意外操作或低效，这可能是谨慎的做法。 LangChain 代理支持内置的 Human in the Loop中间件，以增加对代理工具调用的监督。让我们配置代理，以便在调用 sql_db_query 工具时暂停以供人工审查：

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware 
from langgraph.checkpoint.memory import InMemorySaver 


agent = create_agent(
    model,
    tools,
    system_prompt=system_prompt,
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={"sql_db_query": True},
            description_prefix="Tool execution pending approval",
        ),
    ],
    checkpointer=InMemorySaver(),
)

我们为代理添加了一个检查点，以允许执行暂停和恢复。有关此内容以及可用中间件配置的详细信息，请参阅 Human in the Loop指南。

运行代理时，它现在将在执行 sql_db_query 工具之前暂停以供审查：

question = "Which genre on average has the longest tracks?"
config = {"configurable": {"thread_id": "1"}}

for step in agent.stream(
    {"messages": [{"role": "user", "content": question}]},
    config,
    stream_mode="values",
):
    if "__interrupt__" in step:
        print("INTERRUPTED:")
        interrupt = step["__interrupt__"][0]
        for request in interrupt.value["action_requests"]:
            print(request["description"])
    elif "messages" in step:
        step["messages"][-1].pretty_print()
    else:
        pass

...

INTERRUPTED:
Tool execution pending approval

Tool: sql_db_query
Args: {'query': 'SELECT g.Name AS Genre, AVG(t.Milliseconds) AS AvgTrackLength FROM Track t JOIN Genre g ON t.GenreId = g.GenreId GROUP BY g.Name ORDER BY AvgTrackLength DESC LIMIT 1;'}

我们可以恢复执行，在这种情况下接受查询，使用 Command：

from langgraph.types import Command 

for step in agent.stream(
    Command(resume={"decisions": [{"type": "approve"}]}),
    config,
    stream_mode="values",
):
    if "messages" in step:
        step["messages"][-1].pretty_print()
    if "__interrupt__" in step:
        print("INTERRUPTED:")
        interrupt = step["__interrupt__"][0]
        for request in interrupt.value["action_requests"]:
            print(request["description"])
    else:
        pass

================================== Ai Message ==================================
Tool Calls:
  sql_db_query (call_7oz86Epg7lYRqi9rQHbZPS1U)
 Call ID: call_7oz86Epg7lYRqi9rQHbZPS1U
  Args:
    query: SELECT Genre.Name, AVG(Track.Milliseconds) AS AvgDuration FROM Track JOIN Genre ON Track.GenreId = Genre.GenreId GROUP BY Genre.Name ORDER BY AvgDuration DESC LIMIT 5;
================================= Tool Message =================================
Name: sql_db_query

[('Sci Fi & Fantasy', 2911783.0384615385), ('Science Fiction', 2625549.076923077), ('Drama', 2575283.78125), ('TV Shows', 2145041.0215053763), ('Comedy', 1585263.705882353)]
================================== Ai Message ==================================

The genre with the longest average track length is "Sci Fi & Fantasy" with an average duration of about 2,911,783 milliseconds, followed by "Science Fiction" and "Drama."

有关详细信息，请参阅 Human in the Loop指南。

后续步骤

如需更深入的自定义，请查看本教程，了解如何直接使用 LangGraph 原语实现 SQL 代理。

将这些文档通过 MCP 连接到 Claude、VSCode 等，以获取实时答案。

在 GitHub 上编辑此页面或提交问题。

Tutorials

Conceptual overviews

Additional resources

概述

概念

设置

安装

LangSmith

1. 选择一个 LLM

2. 配置数据库

3. 添加用于数据库交互的工具

4. 使用 `create_agent`

5. 运行代理

（可选）使用 Studio

6. 实现Human in the Loop审查

后续步骤

​概述

​概念

​设置

​安装

​LangSmith

​1. 选择一个 LLM

​2. 配置数据库

​3. 添加用于数据库交互的工具

​4. 使用 create_agent

​5. 运行代理

​（可选）使用 Studio

​6. 实现Human in the Loop审查

​后续步骤

概述

概念

设置

安装

LangSmith

1. 选择一个 LLM

2. 配置数据库

3. 添加用于数据库交互的工具

4. 使用 `create_agent`

5. 运行代理

（可选）使用 Studio

6. 实现Human in the Loop审查

后续步骤