Qdrant

High-performance vector database for semantic search and RAG applications — GPU-accelerated indexing

Qdrant is an open-source, production-ready vector database written in Rust. It delivers fast approximate nearest neighbor (ANN) search across billions of vectors with advanced filtering, payload indexing, and multi-vector support. It's the backbone of many production RAG (Retrieval-Augmented Generation) pipelines and semantic search applications.

GitHub: qdrant/qdrant — 22K+ ⭐

---

Why Qdrant?

Feature	Qdrant	Pinecone	Weaviate	Chroma
Open source	✅	❌	✅	✅
Rust performance	✅	—	❌ Go	❌ Python
Filtering at query time	✅ Advanced	✅ Basic	✅	✅ Basic
Multi-vector	✅	❌	✅	❌
Disk-based HNSW	✅	✅	✅	❌
Payload indexing	✅	Limited	✅	Limited
gRPC + REST	✅ Both	✅ REST	✅	REST
Self-hosted	✅	❌ Cloud only	✅	✅

Qdrant is written in Rust — delivering C-level performance with memory safety. Benchmark tests show Qdrant is consistently 1.5–3x faster than Python-based alternatives like Chroma for high-load scenarios.

---

Key Use Cases

• RAG (Retrieval-Augmented Generation) — find relevant context for LLM prompts

• Semantic search — search by meaning, not just keywords

• Recommendation systems — find similar items by embedding similarity

• Duplicate detection — identify near-duplicate content

• Anomaly detection — find vectors far from cluster centers

• Image/audio similarity search — multimodal retrieval

---

Prerequisites

• Clore.ai account with GPU rental

• Basic familiarity with REST APIs or Python

• Your embedding model of choice (OpenAI, SentenceTransformers, etc.)

---

Step 1 — Rent a Server on Clore.ai

Qdrant is primarily CPU/RAM-bound for serving, but benefits from GPU when:

• Generating embeddings alongside serving (embedding model on same server)

• Large-scale batch indexing operations

1. Go to clore.ai → Marketplace

2. For embeddings + serving combo: RTX 3090/4090 with 32GB+ RAM

3. For serving only: CPU-optimized server with fast NVMe storage

Memory Planning:

• Each float32 vector with 1536 dimensions = 6KB

• 1 million vectors = ~6GB RAM

• 10 million vectors = ~60GB RAM

• Enable on-disk storage for very large collections

---

Step 2 — Deploy Qdrant Container

Docker Image:

qdrant/qdrant:latest

Ports:

22
6333
6334

• Port 6333: REST API (HTTP)

• Port 6334: gRPC API (higher performance for bulk operations)

Environment Variables:

QDRANT__SERVICE__HTTP_PORT=6333
QDRANT__SERVICE__GRPC_PORT=6334
QDRANT__LOG_LEVEL=INFO
QDRANT__STORAGE__STORAGE_PATH=/qdrant/storage

Volume/Persistent Storage: Mount /qdrant/storage for data persistence. Without this, data is lost on container restart.

---

Step 3 — Verify Qdrant is Running

ssh root@<server-ip> -p <ssh-port>

# Check Qdrant is running
curl http://localhost:6333/

# Expected response:
# {"title":"qdrant - vector search engine","version":"..."}

# Check health
curl http://localhost:6333/healthz

# Check cluster info
curl http://localhost:6333/cluster

---

Step 4 — Install Python Client

# Install Qdrant Python client and embedding tools
pip install qdrant-client sentence-transformers openai numpy

# Verify connection
python3 << 'EOF'
from qdrant_client import QdrantClient

client = QdrantClient("localhost", port=6333)
print(f"Qdrant connected: {client.get_collections()}")
EOF

---

Step 5 — Create a Collection

A collection is a named group of vectors with a fixed dimensionality.

from qdrant_client import QdrantClient
from qdrant_client.models import (
    Distance,
    VectorParams,
    HnswConfigDiff,
    OptimizersConfigDiff,
    QuantizationConfig,
    ScalarQuantizationConfig,
    ScalarType
)

client = QdrantClient("localhost", port=6333)

# Create collection for OpenAI text-embedding-3-small (1536 dims)
client.create_collection(
    collection_name="documents",
    vectors_config=VectorParams(
        size=1536,           # Vector dimension (match your embedding model)
        distance=Distance.COSINE,  # Options: COSINE, EUCLID, DOT
        on_disk=False        # Set True for very large collections
    ),
    hnsw_config=HnswConfigDiff(
        m=16,                # HNSW graph connectivity (higher = better recall, more RAM)
        ef_construct=100,    # Build-time search depth (higher = better quality, slower indexing)
        full_scan_threshold=10000  # Use brute force below this count
    ),
    optimizers_config=OptimizersConfigDiff(
        indexing_threshold=20000  # Start HNSW indexing after this many vectors
    ),
    quantization_config=QuantizationConfig(
        scalar=ScalarQuantizationConfig(
            type=ScalarType.INT8,  # Compress vectors to INT8 (4x memory reduction)
            quantile=0.99,
            always_ram=True        # Keep quantized index in RAM
        )
    )
)

print("Collection created!")
print(client.get_collection("documents"))

Collection for SentenceTransformers (384 dims)

client.create_collection(
    collection_name="embeddings_384",
    vectors_config=VectorParams(
        size=384,              # all-MiniLM-L6-v2 output size
        distance=Distance.COSINE
    )
)

---

Step 6 — Index Documents

With OpenAI Embeddings

from qdrant_client import QdrantClient
from qdrant_client.models import PointStruct
from openai import OpenAI
import uuid

client = QdrantClient("localhost", port=6333)
openai_client = OpenAI(api_key="your-openai-api-key")

def get_embeddings(texts: list[str], batch_size: int = 100) -> list[list[float]]:
    """Generate embeddings in batches."""
    all_embeddings = []
    for i in range(0, len(texts), batch_size):
        batch = texts[i:i + batch_size]
        response = openai_client.embeddings.create(
            model="text-embedding-3-small",
            input=batch
        )
        all_embeddings.extend([e.embedding for e in response.data])
    return all_embeddings

# Sample documents
documents = [
    {
        "id": str(uuid.uuid4()),
        "text": "Qdrant is a vector database built in Rust for high performance.",
        "source": "documentation",
        "category": "database",
        "year": 2024
    },
    {
        "id": str(uuid.uuid4()),
        "text": "Machine learning models convert text to dense vector representations.",
        "source": "article",
        "category": "ml",
        "year": 2023
    },
    # Add more documents...
]

# Generate embeddings
texts = [doc["text"] for doc in documents]
embeddings = get_embeddings(texts)

# Upsert into Qdrant
points = [
    PointStruct(
        id=str(uuid.uuid4()),
        vector=embedding,
        payload={
            "text": doc["text"],
            "source": doc["source"],
            "category": doc["category"],
            "year": doc["year"]
        }
    )
    for doc, embedding in zip(documents, embeddings)
]

client.upsert(
    collection_name="documents",
    points=points,
    wait=True  # Wait for indexing to complete
)

print(f"Indexed {len(points)} documents!")

With SentenceTransformers (Local, GPU-accelerated)

from sentence_transformers import SentenceTransformer
from qdrant_client import QdrantClient
from qdrant_client.models import PointStruct
import torch
import uuid

# Load embedding model on GPU
model = SentenceTransformer("all-MiniLM-L6-v2", device="cuda")

client = QdrantClient("localhost", port=6333)

documents = [
    {"text": "How do I set up Qdrant on a GPU server?", "tag": "setup"},
    {"text": "Vector databases store high-dimensional embeddings for similarity search.", "tag": "concept"},
    {"text": "HNSW algorithm provides approximate nearest neighbor search.", "tag": "algorithm"},
    # ... more documents
]

# GPU-accelerated batch encoding
texts = [doc["text"] for doc in documents]
embeddings = model.encode(
    texts,
    batch_size=256,       # Large batch size for GPU efficiency
    show_progress_bar=True,
    normalize_embeddings=True  # Normalize for cosine similarity
)

# Index in Qdrant
points = [
    PointStruct(
        id=str(uuid.uuid4()),
        vector=embedding.tolist(),
        payload=doc
    )
    for doc, embedding in zip(documents, embeddings)
]

# Batch upsert (more efficient)
BATCH_SIZE = 1000
for i in range(0, len(points), BATCH_SIZE):
    batch = points[i:i + BATCH_SIZE]
    client.upsert(collection_name="embeddings_384", points=batch)
    print(f"Indexed {min(i + BATCH_SIZE, len(points))}/{len(points)}")

---

Step 7 — Search and Query

Basic Semantic Search

from qdrant_client import QdrantClient
from sentence_transformers import SentenceTransformer

client = QdrantClient("localhost", port=6333)
model = SentenceTransformer("all-MiniLM-L6-v2", device="cuda")

def search(query: str, limit: int = 5, collection: str = "embeddings_384"):
    # Generate query embedding
    query_vector = model.encode(query, normalize_embeddings=True).tolist()
    
    # Search
    results = client.search(
        collection_name=collection,
        query_vector=query_vector,
        limit=limit,
        with_payload=True,
        with_vectors=False    # Don't return vectors (saves bandwidth)
    )
    
    return results

# Test search
results = search("vector database performance")
for r in results:
    print(f"Score: {r.score:.4f} | {r.payload['text'][:100]}")

Filtered Search (Metadata + Vector)

from qdrant_client.models import Filter, FieldCondition, MatchValue, Range

# Search with metadata filters
results = client.search(
    collection_name="documents",
    query_vector=query_vector,
    query_filter=Filter(
        must=[
            FieldCondition(
                key="category",
                match=MatchValue(value="database")
            ),
            FieldCondition(
                key="year",
                range=Range(gte=2023)  # Year >= 2023
            )
        ]
    ),
    limit=10,
    with_payload=True
)

Batch/Multi-Query Search

from qdrant_client.models import SearchRequest

queries = [
    "how to install vector database",
    "machine learning inference optimization",
    "RAG pipeline architecture"
]

query_vectors = model.encode(queries, normalize_embeddings=True)

# Batch search (one API call for all queries)
results = client.search_batch(
    collection_name="embeddings_384",
    requests=[
        SearchRequest(
            vector=vec.tolist(),
            limit=5,
            with_payload=True
        )
        for vec in query_vectors
    ]
)

for query, res in zip(queries, results):
    print(f"\nQuery: {query}")
    for r in res:
        print(f"  {r.score:.3f}: {r.payload['text'][:80]}")

---

Step 8 — Build a RAG Pipeline

from qdrant_client import QdrantClient
from sentence_transformers import SentenceTransformer
from openai import OpenAI

# Initialize clients
qdrant = QdrantClient("localhost", port=6333)
embedder = SentenceTransformer("all-MiniLM-L6-v2", device="cuda")
llm = OpenAI(api_key="your-openai-key")

def rag_query(question: str, n_context: int = 5) -> str:
    # Step 1: Embed the question
    query_vector = embedder.encode(question, normalize_embeddings=True).tolist()
    
    # Step 2: Retrieve relevant context from Qdrant
    search_results = qdrant.search(
        collection_name="documents",
        query_vector=query_vector,
        limit=n_context,
        with_payload=True
    )
    
    # Step 3: Build context string
    context = "\n\n".join([
        f"[Source: {r.payload.get('source', 'unknown')}]\n{r.payload['text']}"
        for r in search_results
        if r.score > 0.5  # Filter low-confidence results
    ])
    
    # Step 4: Generate answer with LLM
    response = llm.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {
                "role": "system",
                "content": "Answer questions based on the provided context. Be concise and accurate."
            },
            {
                "role": "user",
                "content": f"Context:\n{context}\n\nQuestion: {question}"
            }
        ],
        temperature=0.1
    )
    
    return response.choices[0].message.content

# Test RAG pipeline
answer = rag_query("What is Qdrant and how does it work?")
print(answer)

---

Step 9 — Monitor and Manage Collections

# Collection statistics
info = client.get_collection("documents")
print(f"Vectors count: {info.vectors_count:,}")
print(f"Points count: {info.points_count:,}")
print(f"Indexed vectors: {info.indexed_vectors_count:,}")
print(f"Status: {info.status}")
print(f"Disk usage: {info.disk_data_size / 1024 / 1024:.1f} MB")

# List all collections
collections = client.get_collections()
for c in collections.collections:
    print(f" - {c.name}")

# Delete points by filter
client.delete(
    collection_name="documents",
    points_selector=Filter(
        must=[FieldCondition(key="source", match=MatchValue(value="old_source"))]
    )
)

# Optimize collection (force index build)
client.update_collection(
    collection_name="documents",
    optimizer_config=OptimizersConfigDiff(indexing_threshold=0)  # Force immediate indexing
)

---

Troubleshooting

Connection Refused

# Check Qdrant is running
docker ps | grep qdrant
# Or check the process
ps aux | grep qdrant

# Check ports are open
curl http://localhost:6333/
netstat -tlnp | grep 6333

Slow Search Performance

# Optimize HNSW parameters for better recall
client.update_collection(
    collection_name="documents",
    hnsw_config=HnswConfigDiff(ef=128)  # Increase search-time ef (default 100)
)

# Use INT8 quantization to fit more vectors in RAM

High Memory Usage

# Enable on-disk storage for large collections
client.create_collection(
    collection_name="large_collection",
    vectors_config=VectorParams(
        size=1536,
        distance=Distance.COSINE,
        on_disk=True  # Store vectors on disk instead of RAM
    )
)

---

REST API Quick Reference

# List collections
curl http://localhost:6333/collections

# Create collection
curl -X PUT http://localhost:6333/collections/my_collection \
    -H "Content-Type: application/json" \
    -d '{"vectors": {"size": 384, "distance": "Cosine"}}'

# Count points
curl http://localhost:6333/collections/my_collection/points/count

# Search
curl -X POST http://localhost:6333/collections/my_collection/points/search \
    -H "Content-Type: application/json" \
    -d '{
        "vector": [0.1, 0.2, ...],
        "limit": 5,
        "with_payload": true
    }'

# Delete collection
curl -X DELETE http://localhost:6333/collections/my_collection

---

Cost Estimation on Clore.ai

Setup	Server	Monthly Cost	Capacity
Small RAG	RTX 3090, 32GB RAM	~$60–80	~5M vectors
Medium search	RTX 4090, 64GB RAM	~$120–150	~15M vectors
Large scale	A100, 128GB RAM	~$250–350	~30M vectors

---

Additional Resources

• Qdrant Documentation

• Qdrant GitHub

• Qdrant Python Client

• Qdrant Examples

• Vector Database Benchmarks

• Sentence Transformers

---

Qdrant on Clore.ai gives you a self-hosted, high-performance vector database without the per-query costs of Pinecone or Weaviate Cloud. Perfect for RAG pipelines processing millions of documents.

---

Clore.ai GPU Recommendations

Use Case	Recommended GPU	Est. Cost on Clore.ai
Development/Testing	RTX 3090 (24GB)	~$0.12/gpu/hr
Production Vector Search	RTX 3090 (24GB)	~$0.12/gpu/hr
High-throughput Embedding	RTX 4090 (24GB)	~$0.70/gpu/hr

💡 All examples in this guide can be deployed on Clore.ai GPU servers. Browse available GPUs and rent by the hour — no commitments, full root access.