How to beginner · 3 min read

How to use Vertex AI embeddings with LangChain

Q: How to use Vertex AI embeddings with LangChain

Use the vertexai SDK to generate embeddings with a Vertex AI embedding model, then wrap those embeddings with OpenAIEmbeddings or a custom LangChain embeddings class. Initialize LangChain vectorstores like FAISS with these embeddings for semantic search. This enables seamless integration of Vertex AI embeddings in LangChain pipelines.

Quick answer

Use the vertexai SDK to generate embeddings with a Vertex AI embedding model, then wrap those embeddings with OpenAIEmbeddings or a custom LangChain embeddings class. Initialize LangChain vectorstores like FAISS with these embeddings for semantic search. This enables seamless integration of Vertex AI embeddings in LangChain pipelines.

PREREQUISITES

Python 3.8+
Google Cloud project with Vertex AI enabled
Google Cloud SDK installed and authenticated
pip install vertexai langchain langchain_community faiss-cpu

Setup

Install the required Python packages and authenticate your Google Cloud environment to use Vertex AI.

Install packages: vertexai, langchain, langchain_community, and faiss-cpu for vector search.
Authenticate with Google Cloud using gcloud auth application-default login or set GOOGLE_APPLICATION_CREDENTIALS environment variable.

bash

pip install vertexai langchain langchain_community faiss-cpu

Step by step

This example shows how to generate embeddings from Vertex AI and use them with LangChain's FAISS vectorstore for semantic search.

python

import os
import vertexai
from vertexai.language_models import TextEmbeddingModel
from langchain_community.vectorstores import FAISS
from langchain.schema import Document

# Initialize Vertex AI SDK
vertexai.init(project=os.environ['GOOGLE_CLOUD_PROJECT'], location='us-central1')

# Load Vertex AI embedding model
embedding_model = TextEmbeddingModel.from_pretrained('textembedding-gecko@001')

# Define a wrapper class to use Vertex AI embeddings with LangChain
class VertexAIEmbeddings:
    def embed_documents(self, texts):
        embeddings = embedding_model.get_embeddings(texts)
        return [e.values for e in embeddings]

    def embed_query(self, text):
        embedding = embedding_model.get_embeddings([text])[0]
        return embedding.values

# Sample documents
texts = [
    "LangChain enables easy LLM integrations.",
    "Vertex AI provides powerful embedding models.",
    "FAISS is a fast vector search library."
]

# Create LangChain documents
docs = [Document(page_content=t) for t in texts]

# Instantiate embeddings
embeddings = VertexAIEmbeddings()

# Create FAISS vectorstore from documents
vectorstore = FAISS.from_documents(docs, embeddings)

# Query vectorstore
query = "What library helps with vector search?"
query_embedding = embeddings.embed_query(query)

# Retrieve top 2 relevant documents
results = vectorstore.similarity_search_by_vector(query_embedding, k=2)

for i, doc in enumerate(results, 1):
    print(f"Result {i}: {doc.page_content}")

output

Result 1: FAISS is a fast vector search library.
Result 2: LangChain enables easy LLM integrations.

Common variations

Use async Vertex AI calls by adapting vertexai SDK methods with asyncio.
Swap FAISS with other LangChain vectorstores like Chroma or Weaviate.
Change embedding model by specifying a different Vertex AI embedding model name in TextEmbeddingModel.from_pretrained().

Troubleshooting

If you get authentication errors, ensure your GOOGLE_APPLICATION_CREDENTIALS env var points to a valid service account JSON key with Vertex AI permissions.
If embeddings are empty or errors occur, verify the model name textembedding-gecko@001 is available in your region.
For vectorstore errors, confirm faiss-cpu is installed and compatible with your Python version.

✅

Key Takeaways

Use the vertexai SDK to generate embeddings with Vertex AI models.
Wrap Vertex AI embeddings in a LangChain-compatible class for vectorstores.
Integrate with LangChain vectorstores like FAISS for semantic search.
Authenticate properly with Google Cloud to avoid permission issues.
You can switch embedding models or vectorstores easily within this pattern.

Verified 2026-04 · textembedding-gecko@001

Verify ↗