Refactoring Environment Variable Management in LangChain Projects

chap10 Project

chap10 is part of a LangChain learning series covering dotenv-based environment variable management and project structure. It spans 10 commits made in February 2025.

Problem: Hardcoded API Keys

# Bad example
llm = ChatOpenAI(
    api_key="sk-xxxxx...",  # Never do this!
    model="gpt-4"
)

Solution: The dotenv Pattern

flowchart TD
    A[.env 파일] --> B[dotenv 로드]
    B --> C[os.environ]
    C --> D[LangChain 자동 인식]

    E[.env.example] --> F[팀원 공유용]
    G[.gitignore] --> H[.env 제외]

from dotenv import load_dotenv
import os

load_dotenv()

# LangChain automatically picks up OPENAI_API_KEY
llm = ChatOpenAI(model="gpt-4")

.env File Structure

OPENAI_API_KEY=sk-...
LANGCHAIN_API_KEY=ls-...
LANGCHAIN_TRACING_V2=true
LANGCHAIN_PROJECT=chap10
TAVILY_API_KEY=tvly-...

LangSmith Tracing

With just the environment variables set, tracing data is sent to LangSmith automatically:

# Just set these in .env — that's all it takes
# LANGCHAIN_TRACING_V2=true
# LANGCHAIN_API_KEY=ls-...
# LANGCHAIN_PROJECT=chap10

# No additional configuration needed in code
chain = prompt | llm | parser
result = chain.invoke({"input": "test"})
# → Traces appear in the LangSmith dashboard

Takeaways

1. Never hardcode API keys in source code
2. Use .env.example to document required variables
3. LangSmith tracing is essential for debugging
4. Follow the naming conventions the library expects for environment variables