Elsai Embeddings

Package: elsai-embeddings  v0.2.0

Generates vector embeddings for text using Azure OpenAI and Amazon Bedrock models. Embeddings are the foundation for semantic search and vector store retrieval.

Installation

pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-embeddings/ elsai-embeddings==0.2.0

Requirements: Python >= 3.9

---

Available models

Class	Import path	Provider
AzureOpenAIEmbeddingModel	elsai_embeddings.azure_embeddings	Azure OpenAI
BedrockEmbeddings	elsai_embeddings.bedrock_embeddings	Amazon Bedrock

---

AzureOpenAIEmbeddingModel

Generates vector embeddings using an Azure OpenAI deployment.

from elsai_embeddings.azure_embeddings import AzureOpenAIEmbeddingModel

embedding_model = AzureOpenAIEmbeddingModel(
    model="text-embedding-ada-002",
    azure_api_key="your-azure-api-key",
    azure_api_version="2023-05-15",
    azure_deployment="your-deployment-name",
    azure_endpoint="https://your-resource.openai.azure.com/",
)

# Embed a single query
vector = embedding_model.embed_query("What is machine learning?")

# Embed multiple documents
vectors = embedding_model.embed_documents([
    "Machine learning is a subset of AI.",
    "Deep learning uses neural networks.",
])

# Access the underlying model instance
model_instance = embedding_model.get_embedding_model()

Constructor parameters:

Parameter	Description
model	Model name identifier (e.g. "text-embedding-ada-002")
azure_api_key	Azure OpenAI API key
azure_api_version	API version (e.g. "2023-05-15")
azure_deployment	Azure deployment name
azure_endpoint	Azure OpenAI service endpoint URL

Methods:

Method	Description
embed_query(text)	Returns a single embedding vector for the given text string
embed_documents(list)	Returns a list of embedding vectors for multiple documents
get_embedding_model()	Returns the underlying model instance

Environment variables: AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT, OPENAI_API_VERSION, AZURE_EMBEDDING_DEPLOYMENT_NAME

---

BedrockEmbeddings

Generates vector embeddings using Amazon Bedrock models, including Amazon Titan and Cohere variants.

Supported models:

Model name	Description
titan-embed-v2	Amazon Titan Embeddings V2
cohere-multilingual	Cohere multilingual embedding model
cohere-english	Cohere English embedding model

from elsai_embeddings.bedrock_embeddings import BedrockEmbeddings

embedding_model = BedrockEmbeddings(
    aws_access_key="your-aws-access-key",
    aws_secret_key="your-aws-secret-key",
    aws_session_token="your-session-token",  # optional, for temporary credentials
    aws_region="us-east-1",
    model_name="titan-embed-v2",
)

# Embed a single text
vector = embedding_model.embed_text("Explain transformers.")

# Embed multiple texts
vectors = embedding_model.embed_texts([
    "Document one content.",
    "Document two content.",
])

Constructor parameters:

Parameter	Description
aws_access_key	AWS access key ID
aws_secret_key	AWS secret access key
aws_session_token	AWS session token (optional — for temporary credentials)
aws_region	AWS region where Bedrock is available (e.g. "us-east-1")
model_name	Bedrock embedding model to use (see supported models above)
config	Optional configuration object

Methods:

Method	Description
embed_text(text)	Returns a single embedding vector for the given text string
embed_texts(list)	Returns a list of embedding vectors for multiple texts

Environment variables: AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION, AWS_SESSION_TOKEN, BEDROCK_EMBEDDING_MODEL_NAME

---

Using with vector stores

Embedding models integrate directly with elsai-vectordb:

from elsai_embeddings.azure_embeddings import AzureOpenAIEmbeddingModel
from elsai_vectordb.chromadb import ChromaVectorDb

embedding_model = AzureOpenAIEmbeddingModel(
    model="text-embedding-ada-002",
    azure_api_key="your-azure-api-key",
    azure_api_version="2023-05-15",
    azure_deployment="your-deployment-name",
    azure_endpoint="https://your-resource.openai.azure.com/",
)

chroma = ChromaVectorDb(persist_directory="./db")

retriever = chroma.as_retriever(
    collection_name="my_collection",
    embedding_model=embedding_model,
)

---

Version history

Version	Changes
0.2.0	Amazon Bedrock embedding support added
0.1.0	Initial release with Azure OpenAI embeddings