Qdrant Vector Database Cheat Sheet

Updated 2026-05-21

Next Topic: RAG (Retrieval Augmented Generation) Cheat Sheet

Qdrant is a high-performance, open-source vector database written in Rust, designed to store, index, and search high-dimensional embedding vectors at scale. It sits at the heart of modern Retrieval-Augmented Generation (RAG) pipelines, semantic search, and recommendation systems, providing both a REST and a gRPC API with official clients for Python, TypeScript, Go, Rust, Java, and .NET. The core mental model to internalize before reading these tables is that every searchable object in Qdrant is a "Point" — a vector plus an optional JSON payload — grouped into a "Collection", and virtually every performance and accuracy trade-off (quantization, HNSW tuning, sharding, on-disk storage) ultimately reduces to how those points are stored and indexed within a collection.

What This Cheat Sheet Covers

This topic spans 17 focused tables and 165 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Core Data Model — Collections, Points, and PayloadsTable 2: Collection Configuration ParametersTable 3: HNSW Index ConfigurationTable 4: Quantization MethodsTable 5: Payload Indexing and Field TypesTable 6: Filtering ConditionsTable 7: Point OperationsTable 8: Search and Query APITable 9: Sparse Vectors and Hybrid SearchTable 10: Distributed Deployment and ClusteringTable 11: Snapshots and BackupTable 12: Storage, Optimizer, and Memory ManagementTable 13: Security and AuthenticationTable 14: API Interfaces and SDKsTable 15: Framework IntegrationsTable 16: Monitoring and ObservabilityTable 17: Multitenancy Patterns

Table 1: Core Data Model — Collections, Points, and Payloads

The three fundamental primitives — Collection, Point, and Payload — form the entire data model. Understanding what lives where determines how you design schemas, choose indexing strategies, and write queries.

Concept	Example	Description
Collection	`client.create_collection("products", vectors_config=VectorParams(size=768, distance=Distance.COSINE))`	Named set of points all sharing the same vector dimensionality and distance metric; the primary organizational unit in Qdrant.
Point	`PointStruct(id=1, vector=[0.1, 0.9, ...], payload={"name": "shirt"})`	The atomic record: a vector plus an optional JSON payload; identified by a 64-bit unsigned integer or UUID.
Payload	`{"price": 29.99, "category": "apparel", "in_stock": true}`	Arbitrary JSON metadata attached to a point; supports keyword, integer, float, bool, geo, datetime, and uuid types for filtering.
Named vectors	`vectors={"image": [0.1, ...], "text": [0.8, ...]}`	Multiple vectors of different sizes and modalities coexisting in one point, each with an independent distance metric and index config.

Table 1: Core Data Model — Collections, Points, and Payloads

Concept	Example	Description
Collection	`client.create_collection("products", vectors_config=VectorParams(size=768, distance=Distance.COSINE))`	Named set of points all sharing the same vector dimensionality and distance metric; the primary organizational unit in Qdrant.
Point	`PointStruct(id=1, vector=[0.1, 0.9, ...], payload={"name": "shirt"})`	The atomic record: a vector plus an optional JSON payload; identified by a 64-bit unsigned integer or UUID.
Payload	`{"price": 29.99, "category": "apparel", "in_stock": true}`	Arbitrary JSON metadata attached to a point; supports keyword, integer, float, bool, geo, datetime, and uuid types for filtering.
Named vectors	`vectors={"image": [0.1, ...], "text": [0.8, ...]}`	Multiple vectors of different sizes and modalities coexisting in one point, each with an independent distance metric and index config.