{ "@context": "https://schema.org", "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-04-28-google-gemini-langchain-cheatsheet/#webpage", "@type": [ "WebPage", "ArchiveComponent" ], "articleBody": "April 28, 2025 7 minute read [View Code](https://github.com/philschmid/gemini-samples/blob/main/examples/gemini-langchain.ipynb)\n\nThe `langchain-google-genai` provides access to Google's powerful Gemini models directly via the Gemini API & Google AI Studio. Google AI Studio enables rapid prototyping and experimentation, making it an ideal starting point for individual developers. [LangChain](https://python.langchain.com/) is a framework for developing AI applications. The `langchain-google-genai` package connects LangChain with Google's Gemini models. [LangGraph](https://python.langchain.com/docs/langgraph/) is a library for building stateful, multi-actor applications with LLMs.\n\nAll examples use the `gemini-2.0-flash` model. Gemini 2.5 Pro and 2.5 Flash can be used via `gemini-2.5-pro-preview-03-25` and `gemini-2.5-flash-preview-04-17`. All model ids can be found in the [Gemini API docs](https://ai.google.dev/gemini-api/docs/models).\n\nStart for free and get your API key from [Google AI Studio](https://aistudio.google.com/app/apikey).\n\n1. Install the package `langchain-google-genai`\n\nPython\n\n```python\n%pip install langchain-google-genai\n```\n\n2. Set your API key\n\nPython\n\n```python\nimport getpass\nimport os\n \nif \"GOOGLE_API_KEY\" not in os.environ:\n os.environ[\"GOOGLE_API_KEY\"] = getpass.getpass(\"Enter your Google AI API key: \")\n```\n\n3. Get Started with LangChain.\n\n## Google Gemini with LangChain Chat Models\n\nLearn how to use Google Gemini chat models within LangChain for basic text generation and conversation tasks.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n model=\"gemini-2.0-flash\",\n temperature=0,\n max_tokens=None,\n timeout=None,\n max_retries=2,\n)\n \n# Simple invocation\nmessages = [\n (\"system\", \"You are a helpful assistant that translates English to French.\"),\n (\"human\", \"I love programming.\"),\n]\nresponse = llm.invoke(messages)\nprint(response.content) # Output: J'adore la programmation.\n```\n\n### Chain calls with Prompt Template\n\nDiscover how to chain LangChain prompt templates with Gemini models for flexible and dynamic input processing.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.prompts import ChatPromptTemplate\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n model=\"gemini-2.0-flash\",\n temperature=0,\n)\n \nprompt = ChatPromptTemplate.from_messages([\n (\"system\", \"You are a helpful assistant that translates {input_language} to {output_language}.\"),\n (\"human\", \"{input}\"),\n])\n \nchain = prompt | llm\nresult = chain.invoke({\n \"input_language\": \"English\",\n \"output_language\": \"German\",\n \"input\": \"I love programming.\",\n})\nprint(result.content) # Output: Ich liebe Programmieren.\n```\n\n### Image Input\n\nExplore using image inputs (URLs or local files) with multimodal Gemini models in LangChain for vision tasks.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \n# Using an image URL\nmessage_url = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe this image.\"},\n {\"type\": \"image_url\", \"image_url\": \"https://picsum.photos/seed/picsum/200/300\"},\n ]\n)\nresult_url = llm.invoke([message_url])\nprint(result_url.content)\n \n# Using a local image\nlocal_image_path = \"../assets/react.png\"\nwith open(local_image_path, \"rb\") as image_file:\n encoded_image = base64.b64encode(image_file.read()).decode('utf-8')\n \nmessage_local = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe this image.\"},\n {\"type\": \"image_url\", \"image_url\": f\"data:image/png;base64,{encoded_image}\"}\n ]\n)\nresult_local = llm.invoke([message_local])\nprint(result_local.content)\n```\n\n### Audio Input\n\nUnderstand how to provide audio file data to Gemini models via LangChain for audio processing like transcription.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \naudio_file_path = \"../path/to/your/audio/file.mp3\"\naudio_mime_type = \"audio/mpeg\"\n \nwith open(audio_file_path, \"rb\") as audio_file:\n encoded_audio = base64.b64encode(audio_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Transcribe this audio.\"},\n {\"type\": \"media\", \"data\": encoded_audio, \"mime_type\": audio_mime_type}\n ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n```\n\n### Video Input\n\nSee how to utilize video file input with Gemini models in LangChain for video understanding and analysis.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \nvideo_file_path = \"../path/to/your/video/file.mp4\"\nvideo_mime_type = \"video/mp4\"\n \nwith open(video_file_path, \"rb\") as video_file:\n encoded_video = base64.b64encode(video_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe what's happening in this video.\"},\n {\"type\": \"media\", \"data\": encoded_video, \"mime_type\": video_mime_type}\n ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n```\n\n### Image Generation\n\nGenerate images from text prompts using specialized Gemini models integrated with LangChain.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nimport base64\nfrom IPython.display import Image, display\n \n# Initialize model for image generation\nllm = ChatGoogleGenerativeAI(model=\"models/gemini-2.0-flash-exp-image-generation\")\n \nmessage = {\n \"role\": \"user\",\n \"content\": \"Generate an image of a cat wearing a hat.\",\n}\n \nresponse = llm.invoke(\n [message],\n generation_config=dict(response_modalities=[\"TEXT\", \"IMAGE\"]),\n)\n \n# Display the generated image\nimage_base64 = response.content[0].get(\"image_url\").get(\"url\").split(\",\")[-1]\nimage_data = base64.b64decode(image_base64)\ndisplay(Image(data=image_data, width=300))\n```\n\n### Tool Calling/Function Calling\n\nLearn to implement tool calling (function calling) with Gemini models in LangChain to execute custom functions.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.tools import tool\nfrom langchain_core.messages import ToolMessage\n \n# Define a tool\n@tool(description=\"Get the current weather in a given location\")\ndef get_weather(location: str) -> str:\n return \"It's sunny.\"\n \n# Initialize model and bind the tool\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\nllm_with_tools = llm.bind_tools([get_weather])\n \n# Invoke with a query that should trigger the tool\nquery = \"What's the weather in San Francisco?\"\nai_msg = llm_with_tools.invoke(query)\n \n# Access tool calls in the response\nprint(ai_msg.tool_calls)\n \n# Pass tool results back to the model\ntool_message = ToolMessage(\n content=get_weather(*ai_msg.tool_calls[0]['args']), \n tool_call_id=ai_msg.tool_calls[0]['id']\n)\nfinal_response = llm_with_tools.invoke([ai_msg, tool_message])\nprint(final_response.content)\n```\n\n### Built-in Tools (Google Search, Code Execution)\n\nLeverage Gemini's built-in tools like Google Search and Code Execution directly within your LangChain applications.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom google.ai.generativelanguage_v1beta.types import Tool as GenAITool\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \n# Google Search\nsearch_resp = llm.invoke(\n \"When is the next total solar eclipse in US?\",\n tools=[GenAITool(google_search={})],\n)\nprint(search_resp.content)\n \n# Code Execution\ncode_resp = llm.invoke(\n \"What is 2*2, use python\",\n tools=[GenAITool(code_execution={})],\n)\n \nfor c in code_resp.content:\n if isinstance(c, dict):\n if c[\"type\"] == 'code_execution_result':\n print(f\"Code execution result: {c['code_execution_result']}\")\n elif c[\"type\"] == 'executable_code':\n print(f\"Executable code: {c['executable_code']}\")\n else:\n print(c)\n```\n\n### Structured Output\n\nControl Gemini model output to conform to a specific Pydantic schema for reliable structured data extraction in LangChain.\n\nPython\n\n```python\nfrom langchain_core.pydantic_v1 import BaseModel, Field\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Define the desired structure\nclass Person(BaseModel):\n '''Information about a person.'''\n name: str = Field(..., description=\"The person's name\")\n height_m: float = Field(..., description=\"The person's height in meters\")\n \n# Initialize the model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\", temperature=0)\nstructured_llm = llm.with_structured_output(Person)\n \n# Invoke the model with a query asking for structured information\nresult = structured_llm.invoke(\"Who was the 16th president of the USA, and how tall was he in meters?\")\nprint(result) # Output: name='Abraham Lincoln' height_m=1.93\n```\n\n### Token Usage Tracking\n\nTrack token usage for Gemini model calls within LangChain to monitor costs and API consumption.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \nresult = llm.invoke(\"Explain the concept of prompt engineering in one sentence.\")\n \nprint(result.content)\nprint(\"\\nUsage Metadata:\")\nprint(result.usage_metadata)\n```\n\n## Google Gemini Embeddings with LangChain\n\nGenerate powerful text embeddings using Google Gemini models within the LangChain framework for semantic understanding.\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\n \nembeddings = GoogleGenerativeAIEmbeddings(model=\"models/gemini-embedding-exp-03-07\")\n \n# Embed a single query\nvector = embeddings.embed_query(\"hello, world!\")\n \n# Embed multiple documents\nvectors = embeddings.embed_documents([\n \"Today is Monday\",\n \"Today is Tuesday\",\n \"Today is April Fools day\",\n])\n```\n\n### Using with Vector Store\n\nIntegrate Gemini text embeddings with LangChain vector stores for efficient similarity search and information retrieval.\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom langchain_core.vectorstores import InMemoryVectorStore\n \n# Initialize embeddings\nembeddings = GoogleGenerativeAIEmbeddings(model=\"models/gemini-embedding-exp-03-07\")\n \ntext = \"LangChain is the framework for building context-aware reasoning applications\"\n \n# Create vector store and retriever\nvectorstore = InMemoryVectorStore.from_texts([text], embedding=embeddings)\nretriever = vectorstore.as_retriever()\n \n# Retrieve similar documents\nretrieved_documents = retriever.invoke(\"What is LangChain?\")\nprint(retrieved_documents[0].page_content)\n```\n\n### Task Types\n\nOptimize Gemini embedding performance by selecting the appropriate task type (e.g., retrieval, classification) in LangChain.\n\nPython\n\n```python\n%pip install scikit-learn\n```\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom sklearn.metrics.pairwise import cosine_similarity\n \n# Different task types for different use cases\nquery_embeddings = GoogleGenerativeAIEmbeddings(\n model=\"models/gemini-embedding-exp-03-07\", \n task_type=\"RETRIEVAL_QUERY\" # For queries\n)\ndoc_embeddings = GoogleGenerativeAIEmbeddings(\n model=\"models/gemini-embedding-exp-03-07\", \n task_type=\"RETRIEVAL_DOCUMENT\" # For documents\n)\n \n# Compare similarity\nq_embed = query_embeddings.embed_query(\"What is the capital of France?\")\nd_embed = doc_embeddings.embed_documents([\"The capital of France is Paris.\", \"Philipp likes to eat pizza.\"])\n \nfor i, d in enumerate(d_embed):\n similarity = cosine_similarity([q_embed], [d])[0][0]\n print(f\"Document {i+1} similarity: {similarity}\")\n```\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "authors": [], "captured_at": "2026-09-27T19:29:15.293681927+00:00", "category": "programming", "content_html": "

April 28, 2025 7 minute read View Code

\n

The langchain-google-genai provides access to Google's powerful Gemini models directly via the Gemini API & Google AI Studio. Google AI Studio enables rapid prototyping and experimentation, making it an ideal starting point for individual developers. LangChain is a framework for developing AI applications. The langchain-google-genai package connects LangChain with Google's Gemini models. LangGraph is a library for building stateful, multi-actor applications with LLMs.

\n

All examples use the gemini-2.0-flash model. Gemini 2.5 Pro and 2.5 Flash can be used via gemini-2.5-pro-preview-03-25 and gemini-2.5-flash-preview-04-17. All model ids can be found in the Gemini API docs.

\n

Start for free and get your API key from Google AI Studio.

\n
    \n
  1. Install the package langchain-google-genai
    2. \n
\n

Python

\n
%pip install langchain-google-genai\n
\n
    \n
  1. Set your API key
    2. \n
\n

Python

\n
import getpass\nimport os\n \nif "GOOGLE_API_KEY" not in os.environ:\n    os.environ["GOOGLE_API_KEY"] = getpass.getpass("Enter your Google AI API key: ")\n
\n
    \n
  1. Get Started with LangChain.
    2. \n
\n

Google Gemini with LangChain Chat Models

\n

Learn how to use Google Gemini chat models within LangChain for basic text generation and conversation tasks.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n    model="gemini-2.0-flash",\n    temperature=0,\n    max_tokens=None,\n    timeout=None,\n    max_retries=2,\n)\n \n# Simple invocation\nmessages = [\n    ("system", "You are a helpful assistant that translates English to French."),\n    ("human", "I love programming."),\n]\nresponse = llm.invoke(messages)\nprint(response.content)  # Output: J'adore la programmation.\n
\n

Chain calls with Prompt Template

\n

Discover how to chain LangChain prompt templates with Gemini models for flexible and dynamic input processing.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.prompts import ChatPromptTemplate\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n    model="gemini-2.0-flash",\n    temperature=0,\n)\n \nprompt = ChatPromptTemplate.from_messages([\n    ("system", "You are a helpful assistant that translates {input_language} to {output_language}."),\n    ("human", "{input}"),\n])\n \nchain = prompt | llm\nresult = chain.invoke({\n    "input_language": "English",\n    "output_language": "German",\n    "input": "I love programming.",\n})\nprint(result.content)  # Output: Ich liebe Programmieren.\n
\n

Image Input

\n

Explore using image inputs (URLs or local files) with multimodal Gemini models in LangChain for vision tasks.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\n \n# Using an image URL\nmessage_url = HumanMessage(\n    content=[\n        {"type": "text", "text": "Describe this image."},\n        {"type": "image_url", "image_url": "https://picsum.photos/seed/picsum/200/300"},\n    ]\n)\nresult_url = llm.invoke([message_url])\nprint(result_url.content)\n \n# Using a local image\nlocal_image_path = "../assets/react.png"\nwith open(local_image_path, "rb") as image_file:\n    encoded_image = base64.b64encode(image_file.read()).decode('utf-8')\n \nmessage_local = HumanMessage(\n    content=[\n        {"type": "text", "text": "Describe this image."},\n        {"type": "image_url", "image_url": f"data:image/png;base64,{encoded_image}"}\n    ]\n)\nresult_local = llm.invoke([message_local])\nprint(result_local.content)\n
\n

Audio Input

\n

Understand how to provide audio file data to Gemini models via LangChain for audio processing like transcription.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\n \naudio_file_path = "../path/to/your/audio/file.mp3"\naudio_mime_type = "audio/mpeg"\n \nwith open(audio_file_path, "rb") as audio_file:\n    encoded_audio = base64.b64encode(audio_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n    content=[\n        {"type": "text", "text": "Transcribe this audio."},\n        {"type": "media", "data": encoded_audio, "mime_type": audio_mime_type}\n    ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n
\n

Video Input

\n

See how to utilize video file input with Gemini models in LangChain for video understanding and analysis.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\n \nvideo_file_path = "../path/to/your/video/file.mp4"\nvideo_mime_type = "video/mp4"\n \nwith open(video_file_path, "rb") as video_file:\n    encoded_video = base64.b64encode(video_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n    content=[\n        {"type": "text", "text": "Describe what's happening in this video."},\n        {"type": "media", "data": encoded_video, "mime_type": video_mime_type}\n    ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n
\n

Image Generation

\n

Generate images from text prompts using specialized Gemini models integrated with LangChain.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nimport base64\nfrom IPython.display import Image, display\n \n# Initialize model for image generation\nllm = ChatGoogleGenerativeAI(model="models/gemini-2.0-flash-exp-image-generation")\n \nmessage = {\n    "role": "user",\n    "content": "Generate an image of a cat wearing a hat.",\n}\n \nresponse = llm.invoke(\n    [message],\n    generation_config=dict(response_modalities=["TEXT", "IMAGE"]),\n)\n \n# Display the generated image\nimage_base64 = response.content[0].get("image_url").get("url").split(",")[-1]\nimage_data = base64.b64decode(image_base64)\ndisplay(Image(data=image_data, width=300))\n
\n

Tool Calling/Function Calling

\n

Learn to implement tool calling (function calling) with Gemini models in LangChain to execute custom functions.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.tools import tool\nfrom langchain_core.messages import ToolMessage\n \n# Define a tool\n@tool(description="Get the current weather in a given location")\ndef get_weather(location: str) -> str:\n    return "It's sunny."\n \n# Initialize model and bind the tool\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\nllm_with_tools = llm.bind_tools([get_weather])\n \n# Invoke with a query that should trigger the tool\nquery = "What's the weather in San Francisco?"\nai_msg = llm_with_tools.invoke(query)\n \n# Access tool calls in the response\nprint(ai_msg.tool_calls)\n \n# Pass tool results back to the model\ntool_message = ToolMessage(\n    content=get_weather(*ai_msg.tool_calls[0]['args']), \n    tool_call_id=ai_msg.tool_calls[0]['id']\n)\nfinal_response = llm_with_tools.invoke([ai_msg, tool_message])\nprint(final_response.content)\n
\n

Built-in Tools (Google Search, Code Execution)

\n

Leverage Gemini's built-in tools like Google Search and Code Execution directly within your LangChain applications.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\nfrom google.ai.generativelanguage_v1beta.types import Tool as GenAITool\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\n \n# Google Search\nsearch_resp = llm.invoke(\n    "When is the next total solar eclipse in US?",\n    tools=[GenAITool(google_search={})],\n)\nprint(search_resp.content)\n \n# Code Execution\ncode_resp = llm.invoke(\n    "What is 2*2, use python",\n    tools=[GenAITool(code_execution={})],\n)\n \nfor c in code_resp.content:\n    if isinstance(c, dict):\n        if c["type"] == 'code_execution_result':\n            print(f"Code execution result: {c['code_execution_result']}")\n        elif c["type"] == 'executable_code':\n            print(f"Executable code: {c['executable_code']}")\n    else:\n        print(c)\n
\n

Structured Output

\n

Control Gemini model output to conform to a specific Pydantic schema for reliable structured data extraction in LangChain.

\n

Python

\n
from langchain_core.pydantic_v1 import BaseModel, Field\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Define the desired structure\nclass Person(BaseModel):\n    '''Information about a person.'''\n    name: str = Field(..., description="The person's name")\n    height_m: float = Field(..., description="The person's height in meters")\n \n# Initialize the model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)\nstructured_llm = llm.with_structured_output(Person)\n \n# Invoke the model with a query asking for structured information\nresult = structured_llm.invoke("Who was the 16th president of the USA, and how tall was he in meters?")\nprint(result)  # Output: name='Abraham Lincoln' height_m=1.93\n
\n

Token Usage Tracking

\n

Track token usage for Gemini model calls within LangChain to monitor costs and API consumption.

\n

Python

\n
from langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model="gemini-2.0-flash")\n \nresult = llm.invoke("Explain the concept of prompt engineering in one sentence.")\n \nprint(result.content)\nprint("\\nUsage Metadata:")\nprint(result.usage_metadata)\n
\n

Google Gemini Embeddings with LangChain

\n

Generate powerful text embeddings using Google Gemini models within the LangChain framework for semantic understanding.

\n

Python

\n
from langchain_google_genai import GoogleGenerativeAIEmbeddings\n \nembeddings = GoogleGenerativeAIEmbeddings(model="models/gemini-embedding-exp-03-07")\n \n# Embed a single query\nvector = embeddings.embed_query("hello, world!")\n \n# Embed multiple documents\nvectors = embeddings.embed_documents([\n    "Today is Monday",\n    "Today is Tuesday",\n    "Today is April Fools day",\n])\n
\n

Using with Vector Store

\n

Integrate Gemini text embeddings with LangChain vector stores for efficient similarity search and information retrieval.

\n

Python

\n
from langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom langchain_core.vectorstores import InMemoryVectorStore\n \n# Initialize embeddings\nembeddings = GoogleGenerativeAIEmbeddings(model="models/gemini-embedding-exp-03-07")\n \ntext = "LangChain is the framework for building context-aware reasoning applications"\n \n# Create vector store and retriever\nvectorstore = InMemoryVectorStore.from_texts([text], embedding=embeddings)\nretriever = vectorstore.as_retriever()\n \n# Retrieve similar documents\nretrieved_documents = retriever.invoke("What is LangChain?")\nprint(retrieved_documents[0].page_content)\n
\n

Task Types

\n

Optimize Gemini embedding performance by selecting the appropriate task type (e.g., retrieval, classification) in LangChain.

\n

Python

\n
%pip install scikit-learn\n
\n

Python

\n
from langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom sklearn.metrics.pairwise import cosine_similarity\n \n# Different task types for different use cases\nquery_embeddings = GoogleGenerativeAIEmbeddings(\n    model="models/gemini-embedding-exp-03-07", \n    task_type="RETRIEVAL_QUERY"  # For queries\n)\ndoc_embeddings = GoogleGenerativeAIEmbeddings(\n    model="models/gemini-embedding-exp-03-07", \n    task_type="RETRIEVAL_DOCUMENT"  # For documents\n)\n \n# Compare similarity\nq_embed = query_embeddings.embed_query("What is the capital of France?")\nd_embed = doc_embeddings.embed_documents(["The capital of France is Paris.", "Philipp likes to eat pizza."])\n \nfor i, d in enumerate(d_embed):\n    similarity = cosine_similarity([q_embed], [d])[0][0]\n    print(f"Document {i+1} similarity: {similarity}")\n
\n
\n

Thanks for reading! If you have any questions or feedback, please let me know on Twitter or LinkedIn.

\n", "content_markdown": "April 28, 2025 7 minute read [View Code](https://github.com/philschmid/gemini-samples/blob/main/examples/gemini-langchain.ipynb)\n\nThe `langchain-google-genai` provides access to Google's powerful Gemini models directly via the Gemini API & Google AI Studio. Google AI Studio enables rapid prototyping and experimentation, making it an ideal starting point for individual developers. [LangChain](https://python.langchain.com/) is a framework for developing AI applications. The `langchain-google-genai` package connects LangChain with Google's Gemini models. [LangGraph](https://python.langchain.com/docs/langgraph/) is a library for building stateful, multi-actor applications with LLMs.\n\nAll examples use the `gemini-2.0-flash` model. Gemini 2.5 Pro and 2.5 Flash can be used via `gemini-2.5-pro-preview-03-25` and `gemini-2.5-flash-preview-04-17`. All model ids can be found in the [Gemini API docs](https://ai.google.dev/gemini-api/docs/models).\n\nStart for free and get your API key from [Google AI Studio](https://aistudio.google.com/app/apikey).\n\n1. Install the package `langchain-google-genai`\n\nPython\n\n```python\n%pip install langchain-google-genai\n```\n\n2. Set your API key\n\nPython\n\n```python\nimport getpass\nimport os\n \nif \"GOOGLE_API_KEY\" not in os.environ:\n os.environ[\"GOOGLE_API_KEY\"] = getpass.getpass(\"Enter your Google AI API key: \")\n```\n\n3. Get Started with LangChain.\n\n## Google Gemini with LangChain Chat Models\n\nLearn how to use Google Gemini chat models within LangChain for basic text generation and conversation tasks.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n model=\"gemini-2.0-flash\",\n temperature=0,\n max_tokens=None,\n timeout=None,\n max_retries=2,\n)\n \n# Simple invocation\nmessages = [\n (\"system\", \"You are a helpful assistant that translates English to French.\"),\n (\"human\", \"I love programming.\"),\n]\nresponse = llm.invoke(messages)\nprint(response.content) # Output: J'adore la programmation.\n```\n\n### Chain calls with Prompt Template\n\nDiscover how to chain LangChain prompt templates with Gemini models for flexible and dynamic input processing.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.prompts import ChatPromptTemplate\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(\n model=\"gemini-2.0-flash\",\n temperature=0,\n)\n \nprompt = ChatPromptTemplate.from_messages([\n (\"system\", \"You are a helpful assistant that translates {input_language} to {output_language}.\"),\n (\"human\", \"{input}\"),\n])\n \nchain = prompt | llm\nresult = chain.invoke({\n \"input_language\": \"English\",\n \"output_language\": \"German\",\n \"input\": \"I love programming.\",\n})\nprint(result.content) # Output: Ich liebe Programmieren.\n```\n\n### Image Input\n\nExplore using image inputs (URLs or local files) with multimodal Gemini models in LangChain for vision tasks.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \n# Using an image URL\nmessage_url = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe this image.\"},\n {\"type\": \"image_url\", \"image_url\": \"https://picsum.photos/seed/picsum/200/300\"},\n ]\n)\nresult_url = llm.invoke([message_url])\nprint(result_url.content)\n \n# Using a local image\nlocal_image_path = \"../assets/react.png\"\nwith open(local_image_path, \"rb\") as image_file:\n encoded_image = base64.b64encode(image_file.read()).decode('utf-8')\n \nmessage_local = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe this image.\"},\n {\"type\": \"image_url\", \"image_url\": f\"data:image/png;base64,{encoded_image}\"}\n ]\n)\nresult_local = llm.invoke([message_local])\nprint(result_local.content)\n```\n\n### Audio Input\n\nUnderstand how to provide audio file data to Gemini models via LangChain for audio processing like transcription.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \naudio_file_path = \"../path/to/your/audio/file.mp3\"\naudio_mime_type = \"audio/mpeg\"\n \nwith open(audio_file_path, \"rb\") as audio_file:\n encoded_audio = base64.b64encode(audio_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Transcribe this audio.\"},\n {\"type\": \"media\", \"data\": encoded_audio, \"mime_type\": audio_mime_type}\n ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n```\n\n### Video Input\n\nSee how to utilize video file input with Gemini models in LangChain for video understanding and analysis.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.messages import HumanMessage\nimport base64\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \nvideo_file_path = \"../path/to/your/video/file.mp4\"\nvideo_mime_type = \"video/mp4\"\n \nwith open(video_file_path, \"rb\") as video_file:\n encoded_video = base64.b64encode(video_file.read()).decode('utf-8')\n \nmessage = HumanMessage(\n content=[\n {\"type\": \"text\", \"text\": \"Describe what's happening in this video.\"},\n {\"type\": \"media\", \"data\": encoded_video, \"mime_type\": video_mime_type}\n ]\n)\nresponse = llm.invoke([message])\nprint(response.content)\n```\n\n### Image Generation\n\nGenerate images from text prompts using specialized Gemini models integrated with LangChain.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nimport base64\nfrom IPython.display import Image, display\n \n# Initialize model for image generation\nllm = ChatGoogleGenerativeAI(model=\"models/gemini-2.0-flash-exp-image-generation\")\n \nmessage = {\n \"role\": \"user\",\n \"content\": \"Generate an image of a cat wearing a hat.\",\n}\n \nresponse = llm.invoke(\n [message],\n generation_config=dict(response_modalities=[\"TEXT\", \"IMAGE\"]),\n)\n \n# Display the generated image\nimage_base64 = response.content[0].get(\"image_url\").get(\"url\").split(\",\")[-1]\nimage_data = base64.b64decode(image_base64)\ndisplay(Image(data=image_data, width=300))\n```\n\n### Tool Calling/Function Calling\n\nLearn to implement tool calling (function calling) with Gemini models in LangChain to execute custom functions.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom langchain_core.tools import tool\nfrom langchain_core.messages import ToolMessage\n \n# Define a tool\n@tool(description=\"Get the current weather in a given location\")\ndef get_weather(location: str) -> str:\n return \"It's sunny.\"\n \n# Initialize model and bind the tool\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\nllm_with_tools = llm.bind_tools([get_weather])\n \n# Invoke with a query that should trigger the tool\nquery = \"What's the weather in San Francisco?\"\nai_msg = llm_with_tools.invoke(query)\n \n# Access tool calls in the response\nprint(ai_msg.tool_calls)\n \n# Pass tool results back to the model\ntool_message = ToolMessage(\n content=get_weather(*ai_msg.tool_calls[0]['args']), \n tool_call_id=ai_msg.tool_calls[0]['id']\n)\nfinal_response = llm_with_tools.invoke([ai_msg, tool_message])\nprint(final_response.content)\n```\n\n### Built-in Tools (Google Search, Code Execution)\n\nLeverage Gemini's built-in tools like Google Search and Code Execution directly within your LangChain applications.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\nfrom google.ai.generativelanguage_v1beta.types import Tool as GenAITool\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \n# Google Search\nsearch_resp = llm.invoke(\n \"When is the next total solar eclipse in US?\",\n tools=[GenAITool(google_search={})],\n)\nprint(search_resp.content)\n \n# Code Execution\ncode_resp = llm.invoke(\n \"What is 2*2, use python\",\n tools=[GenAITool(code_execution={})],\n)\n \nfor c in code_resp.content:\n if isinstance(c, dict):\n if c[\"type\"] == 'code_execution_result':\n print(f\"Code execution result: {c['code_execution_result']}\")\n elif c[\"type\"] == 'executable_code':\n print(f\"Executable code: {c['executable_code']}\")\n else:\n print(c)\n```\n\n### Structured Output\n\nControl Gemini model output to conform to a specific Pydantic schema for reliable structured data extraction in LangChain.\n\nPython\n\n```python\nfrom langchain_core.pydantic_v1 import BaseModel, Field\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Define the desired structure\nclass Person(BaseModel):\n '''Information about a person.'''\n name: str = Field(..., description=\"The person's name\")\n height_m: float = Field(..., description=\"The person's height in meters\")\n \n# Initialize the model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\", temperature=0)\nstructured_llm = llm.with_structured_output(Person)\n \n# Invoke the model with a query asking for structured information\nresult = structured_llm.invoke(\"Who was the 16th president of the USA, and how tall was he in meters?\")\nprint(result) # Output: name='Abraham Lincoln' height_m=1.93\n```\n\n### Token Usage Tracking\n\nTrack token usage for Gemini model calls within LangChain to monitor costs and API consumption.\n\nPython\n\n```python\nfrom langchain_google_genai import ChatGoogleGenerativeAI\n \n# Initialize model\nllm = ChatGoogleGenerativeAI(model=\"gemini-2.0-flash\")\n \nresult = llm.invoke(\"Explain the concept of prompt engineering in one sentence.\")\n \nprint(result.content)\nprint(\"\\nUsage Metadata:\")\nprint(result.usage_metadata)\n```\n\n## Google Gemini Embeddings with LangChain\n\nGenerate powerful text embeddings using Google Gemini models within the LangChain framework for semantic understanding.\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\n \nembeddings = GoogleGenerativeAIEmbeddings(model=\"models/gemini-embedding-exp-03-07\")\n \n# Embed a single query\nvector = embeddings.embed_query(\"hello, world!\")\n \n# Embed multiple documents\nvectors = embeddings.embed_documents([\n \"Today is Monday\",\n \"Today is Tuesday\",\n \"Today is April Fools day\",\n])\n```\n\n### Using with Vector Store\n\nIntegrate Gemini text embeddings with LangChain vector stores for efficient similarity search and information retrieval.\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom langchain_core.vectorstores import InMemoryVectorStore\n \n# Initialize embeddings\nembeddings = GoogleGenerativeAIEmbeddings(model=\"models/gemini-embedding-exp-03-07\")\n \ntext = \"LangChain is the framework for building context-aware reasoning applications\"\n \n# Create vector store and retriever\nvectorstore = InMemoryVectorStore.from_texts([text], embedding=embeddings)\nretriever = vectorstore.as_retriever()\n \n# Retrieve similar documents\nretrieved_documents = retriever.invoke(\"What is LangChain?\")\nprint(retrieved_documents[0].page_content)\n```\n\n### Task Types\n\nOptimize Gemini embedding performance by selecting the appropriate task type (e.g., retrieval, classification) in LangChain.\n\nPython\n\n```python\n%pip install scikit-learn\n```\n\nPython\n\n```python\nfrom langchain_google_genai import GoogleGenerativeAIEmbeddings\nfrom sklearn.metrics.pairwise import cosine_similarity\n \n# Different task types for different use cases\nquery_embeddings = GoogleGenerativeAIEmbeddings(\n model=\"models/gemini-embedding-exp-03-07\", \n task_type=\"RETRIEVAL_QUERY\" # For queries\n)\ndoc_embeddings = GoogleGenerativeAIEmbeddings(\n model=\"models/gemini-embedding-exp-03-07\", \n task_type=\"RETRIEVAL_DOCUMENT\" # For documents\n)\n \n# Compare similarity\nq_embed = query_embeddings.embed_query(\"What is the capital of France?\")\nd_embed = doc_embeddings.embed_documents([\"The capital of France is Paris.\", \"Philipp likes to eat pizza.\"])\n \nfor i, d in enumerate(d_embed):\n similarity = cosine_similarity([q_embed], [d])[0][0]\n print(f\"Document {i+1} similarity: {similarity}\")\n```\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "dateCreated": "2026-09-27T19:29:15.293681927+00:00", "date_created": "2025-04-28T00:00:00+00:00", "date_modified": "2025-04-28T00:00:00+00:00", "date_published": "2025-04-28T00:00:00+00:00", "description": "A comprehensive cheatsheet on using Google's Gemini within the LangChain, covering chat functionalities with multimodal inputs, tool usage, structured data generation, and text embedding techniques.", "external_url": "https://www.philschmid.de/gemini-langchain-cheatsheet", "git": { "commit": "ebb0941c5c06a2ab2377342d607535262a3e969e", "path": "items/philschmid-de/2025/04/2025-04-28-google-gemini-langchain-cheatsheet.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/ebb0941c5c06a2ab2377342d607535262a3e969e/items/philschmid-de/2025/04/2025-04-28-google-gemini-langchain-cheatsheet.md" }, "id": "urn:aggr:item:f9958d7cda6be9b41bf30788ce9d3d12f4fa85cf", "isBasedOn": { "@id": "https://www.philschmid.de/gemini-langchain-cheatsheet" }, "mainEntity": { "@id": "https://www.philschmid.de/gemini-langchain-cheatsheet", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-04-28-google-gemini-langchain-cheatsheet/#webpage" }, "dateModified": "2025-04-28T00:00:00+00:00", "datePublished": "2025-04-28T00:00:00+00:00", "headline": "Google Gemini LangChain Cheatsheet", "timeRequired": "PT6M", "url": "https://www.philschmid.de/gemini-langchain-cheatsheet", "wordCount": 1304 }, "name": "Archived snapshot: Google Gemini LangChain Cheatsheet", "preview": null, "reading_minutes": 6, "replicated_at": null, "source": "philschmid-de", "source_name": "philschmid.de - RSS feed", "summary": "A comprehensive cheatsheet on using Google's Gemini within the LangChain, covering chat functionalities with multimodal inputs, tool usage, structured data generation, and text embedding techniques.", "tags": [], "temporalCoverage": "2026-09-27T19:29:15.293681927+00:00", "title": "Google Gemini LangChain Cheatsheet", "url": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-04-28-google-gemini-langchain-cheatsheet/", "word_count": 1304 }