LangChain Not Working? Fix Import Errors, Chain Issues & LLM Connection

Deprecated imports / ModuleNotFoundError

Since LangChain v0.2, provider code was split into separate packages. Importing from langchain directly for models and prompts will fail in newer versions:

# ❌ Old (pre-0.2) — will fail in newer versions
from langchain.chat_models import ChatOpenAI
from langchain.prompts import ChatPromptTemplate
from langchain.schema import HumanMessage

# ✅ New (0.2+) — correct imports
from langchain_openai import ChatOpenAI
from langchain_anthropic import ChatAnthropic
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.messages import HumanMessage

Install the required packages separately:

pip install langchain langchain-openai langchain-anthropic langchain-core
# For community integrations (Ollama, Hugging Face, etc.)
pip install langchain-community
# For vector stores
pip install langchain-chroma  # or langchain-pinecone, langchain-qdrant

• langchain-openai: contains ChatOpenAI, OpenAIEmbeddings — install separately even if you have openai.
• langchain-core: base classes like ChatPromptTemplate, StrOutputParser, RunnablePassthrough — always required.
• langchain-community: integrations for Ollama, Hugging Face, FAISS, and hundreds of other services.

LCEL vs legacy chain syntax

LangChain Expression Language (LCEL) uses the | operator to compose chains. It returns Runnable objects — you must call .invoke() on the full chain:

import os
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser

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

# LCEL (recommended) — use | to compose chains
llm = ChatOpenAI(model="gpt-4o")
prompt = ChatPromptTemplate.from_template("Tell me a joke about {topic}")
parser = StrOutputParser()

chain = prompt | llm | parser  # Runnable sequence

# Invoke
result = chain.invoke({"topic": "Python"})
print(result)

# Async
result = await chain.ainvoke({"topic": "Python"})

# Stream
for chunk in chain.stream({"topic": "Python"}):
    print(chunk, end="")

• Legacy LLMChain still works but shows deprecation warnings — migrate to LCEL for new code.
• Don't mix syntaxes: LCEL chains return Runnable, not a direct string — call .invoke() to get output.
• Async: in async code always use ainvoke() — calling invoke() in an async context will block the event loop.

LLM connection errors

Test your LLM connection independently before building a full chain — this isolates whether the issue is the LLM or the chain logic:

# Test your LLM connection independently first
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o")  # reads OPENAI_API_KEY from env

try:
    response = llm.invoke("Hello")
    print(response.content)
except Exception as e:
    print(f"Error: {e}")  # will show auth or connection error clearly

• OpenAI: ensure OPENAI_API_KEY is set; install langchain-openai not langchain[openai] (the extras syntax is deprecated).
• Anthropic: set ANTHROPIC_API_KEY; install langchain-anthropic separately.
• Ollama: run ollama serve first; default base_url is http://localhost:11434.

RAG / retrieval chain not returning results

A complete LCEL RAG chain using a vector store retriever — check each component independently if results are empty:

from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_chroma import Chroma
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough

# Build retriever
vectorstore = Chroma.from_documents(documents, OpenAIEmbeddings())
retriever = vectorstore.as_retriever(search_kwargs={"k": 4})

# Build RAG chain
template = """Answer based on context:
{context}

Question: {question}"""
prompt = ChatPromptTemplate.from_template(template)
llm = ChatOpenAI(model="gpt-4o")

chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | llm
)

result = chain.invoke("What is the main topic?")

• Empty retriever results: check documents were loaded and embeddings created; verify vectorstore has data with vectorstore.similarity_search("test").
• Install vector store package: pip install langchain-chroma or langchain-pinecone or langchain-qdrant — not included in core.
• Context format: if the retriever returns Document objects, you may need to format them: add a lambda docs: "\n".join(d.page_content for d in docs) step.

Pydantic v1 vs v2 conflicts

LangChain migrated to Pydantic v2 in 0.3. If you see errors like "BaseModel is defined in pydantic v1" or unexpected ValidationError after upgrading:

# Check your pydantic version
pip show pydantic

# For LangChain 0.3+, pydantic v2 is required
pip install pydantic>=2.0

# If a dependency requires pydantic v1, use compatibility mode
# In your code:
from pydantic.v1 import BaseModel  # pydantic v1 compat

• langchain-core >= 0.3 requires Pydantic v2. If mixing old community packages, version conflicts are common.
• Fresh virtualenv: if you have hard-to-resolve conflicts, create a fresh virtualenv and install all LangChain packages at once with pip install langchain langchain-openai langchain-anthropic langchain-core.
• Check all versions: run pip show langchain langchain-core langchain-openai pydantic to see the full picture before upgrading.