LlamaIndex Embeddings Langchain Integration

Common errors

• ImportError: cannot import name 'LangchainEmbedding' from 'llama_index'

  cause The `LangchainEmbedding` class is not directly available at the top-level `llama_index` package. It resides within the `llama_index.embeddings.langchain` submodule.
  fix Change your import statement to `from llama_index.embeddings.langchain import LangchainEmbedding`.

• ModuleNotFoundError: No module named 'langchain.embeddings.base'

  cause This error typically means the `langchain` package (or `langchain-community` for newer versions) is not installed or the installed version is incompatible with LlamaIndex's bridge.
  fix Install or update LangChain: `pip install langchain` or `pip install langchain-community`. If the issue persists, try pinning an older, known-compatible version of `langchain` (e.g., `langchain==0.0.153` as seen in older issues) and then gradually upgrade.

• AttributeError: 'HuggingFaceEmbeddings' object has no attribute 'aembed_query'

  cause Some underlying LangChain embedding models might not have asynchronous embedding methods implemented, leading to this error when LlamaIndex attempts to use them asynchronously. The `LangchainEmbedding` wrapper handles this by falling back to synchronous calls, but you might see warnings.
  fix This is often handled gracefully by the `LangchainEmbedding` wrapper, falling back to sync. If you strictly require async, ensure your specific LangChain embedding model explicitly supports `aembed_query` and `aembed_documents` methods.

Warnings

• breaking Both LlamaIndex and LangChain are rapidly evolving libraries. Import paths and class locations within `langchain` (e.g., moving to `langchain-community`) have changed frequently, leading to `ImportError` or `ModuleNotFoundError` if versions are not compatible or imports are not updated.
  Fix: Always consult the latest official LlamaIndex and LangChain documentation for correct import paths and recommended packages. Pin your library versions carefully in `requirements.txt`.
• gotcha The `llama-index-embeddings-langchain` package is merely a wrapper. You *must* install the underlying LangChain package that provides the actual embedding model you intend to use (e.g., `langchain-community` for `HuggingFaceEmbeddings`). Not installing this dependency will lead to runtime errors when the wrapped model is initialized.
  Fix: Ensure that `langchain` and/or `langchain-community` are installed alongside `llama-index-embeddings-langchain` and any specific dependencies for your chosen LangChain embedding model.
• gotcha When using local HuggingFace embedding models via LangChain, ensure the model weights are downloaded correctly and that there are no network issues if it's the first time using a specific model. Indefinite loading times can indicate a problem with model download.
  Fix: Check network connectivity. For `HuggingFaceEmbeddings`, you might need to pre-download the model or ensure your environment has access to Hugging Face Hub. Increasing verbosity or checking logs for download progress can also help.

Install

• pip install llama-index-embeddings-langchain Install Package
• pip install langchain-community Install LangChain Community (for most models)

Imports

• LangchainEmbedding
  wrong:

  from llama_index import LangchainEmbedding

  correct:

  from llama_index.embeddings.langchain import LangchainEmbedding

  The `LangchainEmbedding` wrapper is located in a specific submodule, not directly under `llama_index`.
• HuggingFaceEmbeddings
  wrong:

  from langchain.embeddings import HuggingFaceEmbeddings

  correct:

  from langchain_community.embeddings import HuggingFaceEmbeddings

  With recent LangChain refactorings, many common embedding models moved from `langchain.embeddings` to `langchain_community.embeddings`.

Quickstart

This quickstart demonstrates how to integrate a LangChain embedding model, specifically a HuggingFace one, into LlamaIndex. It involves initializing the LangChain model, wrapping it with `LangchainEmbedding`, and then setting it as the global embedding model for LlamaIndex. This allows any LlamaIndex component (e.g., VectorStoreIndex) to use this embedding model.

import os
from llama_index.embeddings.langchain import LangchainEmbedding
from llama_index.core import Settings
from langchain_community.embeddings import HuggingFaceEmbeddings

# Ensure the necessary LangChain package is installed
# pip install langchain-community

# 1. Initialize a LangChain embedding model
# Using a local model for demonstration, no API key needed
lc_embed_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# 2. Wrap the LangChain embedding model with LlamaIndex's LangchainEmbedding wrapper
embed_model = LangchainEmbedding(lc_embed_model)

# 3. Set the global embedding model for LlamaIndex
Settings.embed_model = embed_model

# 4. Example usage: get an embedding
text = "This is a test sentence for embedding."
embedding = Settings.embed_model.get_text_embedding(text)

print(f"Embedding length: {len(embedding)}")
print(f"First 10 dimensions of embedding: {embedding[:10]}")

# You can also use it directly without setting global settings
# direct_embedding = embed_model.get_text_embedding("Another sentence.")
# print(f"Direct embedding length: {len(direct_embedding)}")