Skip to content

Vector Databases

EmbeddingFramework supports multiple vector database backends, allowing you to store and query embeddings efficiently.

📦 Supported Databases

Below are the supported vector databases with detailed explanations, technical functions, and usage patterns.

1️⃣ ChromaDB

  • Local and persistent vector storage.
  • Ideal for small to medium-scale projects.
  • No external dependencies for basic usage.

Technical Functions: - __init__(persist_directory: str): Initializes the ChromaDB adapter with a local storage path. - add_texts(texts: List[str], embeddings: List[List[float]]): Stores text and corresponding embeddings. - query(query_text: str, top_k: int): Retrieves the top_k most similar entries.

Example:

from embeddingframework.adapters.vector_dbs import ChromaDBAdapter

db = ChromaDBAdapter(persist_directory="./chroma_store")
db.add_texts(["Hello world"], [[0.1, 0.2, 0.3]])
results = db.query("Hello", top_k=1)
print(results)

2️⃣ Milvus

  • High-performance distributed vector database.
  • Suitable for large-scale deployments.
  • Requires Milvus server running.

Technical Functions: - __init__(host: str, port: str): Connects to a Milvus server. - add_texts(texts: List[str], embeddings: List[List[float]]): Inserts data into Milvus. - query(query_text: str, top_k: int): Performs similarity search.

Example:

from embeddingframework.adapters.milvus_adapter import MilvusAdapter

db = MilvusAdapter(host="localhost", port="19530")
db.add_texts(["Hello world"], [[0.1, 0.2, 0.3]])
results = db.query("Hello", top_k=1)
print(results)

3️⃣ Pinecone

  • Fully managed vector database service.
  • Scales automatically with your needs.
  • Requires API key.

Technical Functions: - __init__(api_key: str, index_name: str): Connects to Pinecone with API key and index. - add_texts(texts: List[str], embeddings: List[List[float]]): Inserts vectors into Pinecone. - query(query_text: str, top_k: int): Retrieves similar vectors.

Example:

from embeddingframework.adapters.pinecone_adapter import PineconeAdapter

db = PineconeAdapter(api_key="YOUR_PINECONE_API_KEY", index_name="my-index")
db.add_texts(["Hello world"], [[0.1, 0.2, 0.3]])
results = db.query("Hello", top_k=1)
print(results)

4️⃣ Weaviate

  • Open-source vector search engine.
  • Supports hybrid search (vector + keyword).
  • Can be self-hosted or cloud-hosted.

Technical Functions: - __init__(url: str): Connects to a Weaviate instance. - add_texts(texts: List[str], embeddings: List[List[float]]): Adds data to Weaviate. - query(query_text: str, top_k: int): Performs hybrid or vector search.

Example:

from embeddingframework.adapters.weaviate_adapter import WeaviateAdapter

db = WeaviateAdapter(url="http://localhost:8080")
db.add_texts(["Hello world"], [[0.1, 0.2, 0.3]])
results = db.query("Hello", top_k=1)
print(results)

🔄 Switching Databases

The API is consistent across all adapters, so switching between databases is as simple as changing the adapter import and initialization.