LangChain Google Vertex AI Integration

3.2.2 · active · verified Sun Mar 29

This package provides LangChain integrations for Google Cloud generative models via the Vertex AI Platform. It enables access to foundational models (like Gemini) and third-party models available on Vertex AI Model Garden, along with services like Vector Search. The current version is 3.2.2, with active development and frequent releases to support new features and address issues.

Warnings

deprecated Classes `ChatVertexAI`, `VertexAI`, and `VertexAIEmbeddings` are officially deprecated in `langchain-google-vertexai` version 3.2.0 and will be removed in 4.0.0. [8, 9, 20]
Fix: Migrate to `ChatGoogleGenerativeAI`, `GoogleGenerativeAI`, and `GoogleGenerativeAIEmbeddings` respectively, from the `langchain-google-genai` package for most Gemini API interactions. `langchain-google-vertexai` remains supported for Vertex AI platform-specific features (Model Garden, Vector Search, etc.). [8, 9, 22]
gotcha Authentication to Google Cloud Vertex AI typically requires Application Default Credentials (ADC). Incorrect or missing authentication setup is a common source of errors. [2, 3, 6, 7]
Fix: Ensure you are authenticated via `gcloud auth application-default login` or by setting the `GOOGLE_APPLICATION_CREDENTIALS` environment variable to the path of a service account JSON key file. Project ID and location might also need to be specified in the model constructor or environment variables. [2, 3, 6, 20]
gotcha There can be confusion between `langchain-google-vertexai` and `langchain-google-genai`. While both offer Google LLM integrations, `langchain-google-vertexai` focuses on Vertex AI platform-specific features (e.g., Model Garden, Vector Search, Anthropic models on Vertex AI), whereas `langchain-google-genai` is for direct Google Generative AI (Gemini API) access. [2, 7, 22, 33, 34]
Fix: For general Gemini model access (Gemini API), use `langchain-google-genai`. For integrations requiring Vertex AI platform features or third-party models deployed on Vertex AI, use `langchain-google-vertexai`. Refer to the official LangChain Google documentation for guidance on which package to use. [22]
gotcha When using tools with Gemini models through Vertex AI, certain Zod schema features (e.g., discriminated unions, union types, positive refinements) are not supported or are automatically converted, potentially leading to unexpected behavior or errors. [3, 19]
Fix: Use simple, flat object structures for tool schemas, replace discriminated unions with enums and optional fields, and prefer `min()` over `positive()` for number constraints. Test your tool schemas thoroughly. [3, 19]

Install

pip install langchain-google-vertexai Install stable version

Imports

ChatVertexAI
```
from langchain_google_vertexai import ChatVertexAI
```
The primary chat model class for Vertex AI. As of v3.2.0, this is deprecated in favor of `ChatGoogleGenerativeAI` from `langchain-google-genai` for unified Gemini API access, but remains for Vertex AI platform-specific features. [9, 22]
VertexAI
```
from langchain_google_vertexai import VertexAI
```
The primary LLM class for Vertex AI completion models. As of v3.2.0, this is deprecated in favor of `GoogleGenerativeAI` from `langchain-google-genai` for unified Gemini API access, but remains for Vertex AI platform-specific features. [9, 22]
VertexAIEmbeddings
```
from langchain_google_vertexai import VertexAIEmbeddings
```
The embeddings class for Vertex AI. As of v3.2.0, this is deprecated in favor of `GoogleGenerativeAIEmbeddings` from `langchain-google-genai` for unified Gemini API access, but remains for Vertex AI platform-specific features. [9, 20, 22]
ChatAnthropicVertex
```
from langchain_google_vertexai import ChatAnthropicVertex
```
For integrating with Anthropic's Claude models hosted on Vertex AI. [9]

Quickstart

This quickstart demonstrates how to instantiate and use the `ChatVertexAI` model for text generation. It also shows an example of multimodal input (commented out) using a vision-capable model. Ensure your Google Cloud project and location are configured, and that you are authenticated to Google Cloud, typically via Application Default Credentials (e.g., `gcloud auth application-default login`) or by setting `GOOGLE_APPLICATION_CREDENTIALS`.

import os
from langchain_core.messages import HumanMessage
from langchain_google_vertexai import ChatVertexAI

# Ensure your Google Cloud Project ID and location are set
# or use GOOGLE_APPLICATION_CREDENTIALS for authentication.
# For example, by running 'gcloud auth application-default login'
# os.environ["GOOGLE_CLOUD_PROJECT"] = os.environ.get("GOOGLE_CLOUD_PROJECT", "your-gcp-project-id")
# os.environ["GOOGLE_CLOUD_LOCATION"] = os.environ.get("GOOGLE_CLOUD_LOCATION", "us-central1")

# Initialize the chat model
# Note: ChatVertexAI is deprecated in favor of ChatGoogleGenerativeAI from langchain_google_genai
# for most Gemini models, but can still be used for Vertex AI specific deployments.
llm = ChatVertexAI(model="gemini-pro") # or "gemini-2.5-flash", "chat-bison", etc.

# Invoke the model with a message
response = llm.invoke("What is the capital of France?")
print(response.content)

# Example with multimodal input (requires a vision model like "gemini-pro-vision")
# from langchain_core.messages import HumanMessage
# llm_vision = ChatVertexAI(model="gemini-pro-vision")
# message_with_image = HumanMessage(
#     content=[
#         {"type": "text", "text": "What's in this image?"},
#         {"type": "image_url", "image_url": {"url": "https://picsum.photos/seed/picsum/200/300"}},
#     ]
# )
# response_vision = llm_vision.invoke([message_with_image])
# print(response_vision.content)

view raw JSON →