pip install llama-index-embeddings-voyageai

export VOYAGE_API_KEY="your-api-key-here"

from llama_index.embeddings.voyageai import VoyageEmbedding

# Initialize the VoyageAI Embedding model
embedding_model = VoyageEmbedding(
    model_name="voyage-3.5",
    voyage_api_key="your-api-key",  # Optional if VOYAGE_API_KEY is set
)

# Get a single embedding
embedding = embedding_model.get_text_embedding("Your text here")
print(f"Embedding dimension: {len(embedding)}")

# Get embeddings for multiple texts
texts = ["Text 1", "Text 2", "Text 3"]
embeddings = embedding_model.get_text_embedding_batch(texts)
print(f"Number of embeddings: {len(embeddings)}")

from llama_index.embeddings.voyageai import VoyageEmbedding

embedding_model = VoyageEmbedding(model_name="voyage-3.5")

# Get query embedding (automatically uses input_type="query")
query_embedding = embedding_model.get_query_embedding(
    "What is machine learning?"
)

# Get document embedding (automatically uses input_type="document")
doc_embedding = embedding_model.get_text_embedding("Machine learning is...")

from llama_index.embeddings.voyageai import VoyageEmbedding

embedding_model = VoyageEmbedding(
    model_name="voyage-3.5",
    voyage_api_key="your-api-key",
    truncation=True,  # Enable text truncation
    output_dtype="float",  # Options: "float", "int8", "uint8", "binary", "ubinary"
    output_dimension=512,  # Reduce dimensionality (256, 512, 1024, 2048)
    embed_batch_size=128,  # Batch size for processing
)

# Use general text embedding with custom input type
embedding = embedding_model.get_general_text_embedding(
    "Your text here", input_type="query"
)

from llama_index.embeddings.voyageai import VoyageEmbedding
from io import BytesIO

# Initialize with multimodal model (truncation=True is REQUIRED)
embedding_model = VoyageEmbedding(
    model_name="voyage-multimodal-3",  # or "voyage-multimodal-3.5" for video support
    truncation=True,  # Required for multimodal models
)

# Embed an image from file path (PNG, JPEG, JPG, WEBP, GIF supported)
image_embedding = embedding_model.get_image_embedding("path/to/image.jpg")
print(f"Image embedding dimension: {len(image_embedding)}")  # 1024

# Embed an image from BytesIO
with open("path/to/image.png", "rb") as f:
    image_data = BytesIO(f.read())
    image_embedding = embedding_model.get_image_embedding(image_data)

# The multimodal model also works with text
text_embedding = embedding_model.get_text_embedding("Description of the image")
query_embedding = embedding_model.get_query_embedding(
    "Find images with red color"
)

# Batch text embeddings
batch_embeddings = embedding_model.get_text_embedding_batch(
    ["Image description 1", "Image description 2", "Image description 3"]
)

from llama_index.embeddings.voyageai import VoyageEmbedding

# Initialize with voyage-multimodal-3.5 for video support
embedding_model = VoyageEmbedding(
    model_name="voyage-multimodal-3.5",
    truncation=True,
)

# Embed a single video (max 20MB, supports MP4, MPEG, MOV, AVI, FLV, MPG, WEBM, WMV, 3GP)
video_embedding = embedding_model.get_video_embedding("path/to/video.mp4")
print(f"Video embedding dimension: {len(video_embedding)}")  # 1024

# Embed multiple videos
video_embeddings = embedding_model.get_video_embeddings(
    ["video1.mp4", "video2.mp4", "video3.mp4"]
)

# Async video embedding
video_embedding = await embedding_model.aget_video_embedding(
    "path/to/video.mp4"
)

from llama_index.embeddings.voyageai import VoyageEmbedding

# Initialize with contextual model
embedding_model = VoyageEmbedding(
    model_name="voyage-context-3", output_dtype="float", output_dimension=1024
)

# The model will use contextualized_embed internally
# providing enhanced embeddings with better context understanding
embeddings = embedding_model.get_text_embedding_batch(
    ["First document chunk", "Second document chunk", "Third document chunk"]
)

import asyncio
from llama_index.embeddings.voyageai import VoyageEmbedding


async def get_embeddings_async():
    # Regular text embeddings
    embedding_model = VoyageEmbedding(model_name="voyage-3.5")

    # Get async query embedding
    query_embedding = await embedding_model.aget_query_embedding("Your query")

    # Get async text embeddings
    embeddings = await embedding_model.aget_text_embedding_batch(
        ["Text 1", "Text 2", "Text 3"]
    )

    # For multimodal image embeddings
    multimodal_model = VoyageEmbedding(
        model_name="voyage-multimodal-3",
        truncation=True,  # Required for multimodal
    )
    image_embedding = await multimodal_model.aget_image_embedding(
        "path/to/image.jpg"
    )

    return query_embedding, embeddings, image_embedding


# Run async function
results = asyncio.run(get_embeddings_async())

from llama_index.core import VectorStoreIndex, Settings, Document
from llama_index.embeddings.voyageai import VoyageEmbedding
from llama_index.llms.openai import OpenAI

# Configure LlamaIndex settings
Settings.llm = OpenAI()
Settings.embed_model = VoyageEmbedding(
    model_name="voyage-3.5", voyage_api_key="your-api-key"
)

# Create documents
documents = [
    Document(text="LlamaIndex is a data framework for LLM applications."),
    Document(text="VoyageAI provides state-of-the-art embedding models."),
    Document(text="Embeddings convert text into numerical vectors."),
]

# Create vector index
index = VectorStoreIndex.from_documents(documents)

# Query the index
query_engine = index.as_query_engine(similarity_top_k=2)
response = query_engine.query("What is LlamaIndex?")
print(response)