Skip to content

Vector Databases

Vector databases store and search high-dimensional embeddings, enabling semantic search and RAG applications.

Why Vector Databases?

Traditional databases search by exact matches. Vector databases search by meaning:

Query: "How do I fix authentication errors?"

Traditional DB: Searches for exact keywords
Vector DB: Finds documents about login issues, auth failures, credential problems

How They Work

┌─────────────────────────────────────────────────────────────┐
│                    Vector Database                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│   Text → Embedding Model → Vector [0.23, -0.45, ..., 0.89] │
│                              │                              │
│                              v                              │
│                     ┌───────────────┐                       │
│                     │  Index (HNSW) │                       │
│                     └───────────────┘                       │
│                              │                              │
│   Query → Embedding → Nearest Neighbor Search → Results    │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Database Comparison

Feature ChromaDB Qdrant Milvus pgvector
Ease of use Excellent Good Moderate Good
Performance Good Excellent Excellent Moderate
Scalability Single node Distributed Distributed PostgreSQL limits
Filtering Basic Advanced Advanced SQL
Persistence File/Memory File/Server Server PostgreSQL
GPU support No No Yes No
Best for Development Production Enterprise PostgreSQL users

In This Section

Document Description
ChromaDB Simple, local-first vector database
Qdrant High-performance production database
Milvus Enterprise-scale vector database
pgvector PostgreSQL extension for vectors

Quick Start

ChromaDB (Simplest)

import chromadb
from chromadb.utils import embedding_functions

# Create client
client = chromadb.PersistentClient(path="./chroma_db")

# Use Ollama embeddings
ollama_ef = embedding_functions.OllamaEmbeddingFunction(
    model_name="nomic-embed-text",
    url="http://localhost:11434/api/embeddings"
)

# Create collection
collection = client.get_or_create_collection(
    name="documents",
    embedding_function=ollama_ef
)

# Add documents
collection.add(
    documents=["Machine learning is...", "Deep learning uses..."],
    ids=["doc1", "doc2"],
    metadatas=[{"source": "ml.pdf"}, {"source": "dl.pdf"}]
)

# Search
results = collection.query(
    query_texts=["What is neural network?"],
    n_results=2
)

Qdrant (Production)

from qdrant_client import QdrantClient
from qdrant_client.models import VectorParams, Distance, PointStruct

# Connect to Qdrant
client = QdrantClient(host="localhost", port=6333)

# Create collection
client.create_collection(
    collection_name="documents",
    vectors_config=VectorParams(size=768, distance=Distance.COSINE)
)

# Add vectors
client.upsert(
    collection_name="documents",
    points=[
        PointStruct(
            id=1,
            vector=embedding_model.embed("Machine learning is..."),
            payload={"source": "ml.pdf", "page": 1}
        )
    ]
)

# Search
results = client.search(
    collection_name="documents",
    query_vector=embedding_model.embed("What is ML?"),
    limit=5
)

Choosing a Database

Use ChromaDB when:

  • Prototyping and development
  • Single-user applications
  • Small to medium datasets (< 1M vectors)
  • You want the simplest setup

Use Qdrant when:

  • Production deployments
  • Need advanced filtering
  • Distributed/replicated setup
  • High query throughput

Use Milvus when:

  • Enterprise scale (billions of vectors)
  • GPU acceleration needed
  • Multi-tenancy required
  • Kubernetes-native deployment

Use pgvector when:

  • Already using PostgreSQL
  • Need SQL joins with vector search
  • Transactional consistency required
  • Moderate vector workloads

Embedding Models

All databases work with these local embedding models:

Model Dimensions Speed Quality
nomic-embed-text 768 Fast Good
mxbai-embed-large 1024 Medium Better
all-minilm 384 Very fast Basic
bge-m3 1024 Medium Best multilingual 
# Pull models with Ollama
ollama pull nomic-embed-text
ollama pull mxbai-embed-large

See Also