Qdrant

Overview

Qdrant is a vector similarity search engine written in Rust that provides high performance, rich filtering, and advanced features like payload-based filtering and hybrid search.

Setup

Local Docker
Qdrant Cloud
Self-Hosted

Run Qdrant locally with Docker:

docker run -p 6333:6333 qdrant/qdrant

Then in Flowise:

Qdrant Server URL: http://localhost:6333
Collection Name: my-collection

Sign up at Qdrant Cloud
Create a cluster and get your URL and API key
Add Qdrant API credential in Flowise
Configure:

Qdrant Server URL: https://xyz-example.cloud.qdrant.io
Credential: Qdrant API Key
Collection Name: my-collection

Install Qdrant server:

# Download binary
wget https://github.com/qdrant/qdrant/releases/download/v1.7.0/qdrant

# Run
./qdrant

Or use Docker Compose for production.

Configuration

Required Parameters

qdrantServerUrl

string

required

URL of your Qdrant instance:

Local: http://localhost:6333
Cloud: https://your-cluster.cloud.qdrant.io

qdrantCollection

string

required

Name of the collection to use (created automatically if doesn’t exist)

embeddings

Embeddings

required

Embedding model for vector generation

Optional Parameters

credential

Qdrant API key credential (required for Qdrant Cloud)

document

Document[]

Documents to upsert into the collection

recordManager

RecordManager

Track indexed documents to prevent duplication

qdrantVectorDimension

number

default:1536

Vector dimension size (must match embedding model):

OpenAI ada-002: 1536
OpenAI text-embedding-3-small: 1536
OpenAI text-embedding-3-large: 3072

qdrantSimilarity

string

default:"Cosine"

Distance metric for similarity:

Cosine - Normalized similarity (recommended)
Euclid - Euclidean distance
Dot - Dot product

contentPayloadKey

string

default:"content"

Key for storing document content in payload

metadataPayloadKey

string

default:"metadata"

Key for storing document metadata in payload

batchSize

number

Batch size for upserting vectors. Useful for large datasets:

batchSize: 100  // Process 100 vectors at a time

qdrantCollectionConfiguration

json

Advanced collection configuration (see Qdrant docs):

{
  "optimizers_config": {
    "indexing_threshold": 10000
  }
}

qdrantFilter

json

Filter search results by payload:

{
  "must": [
    { "key": "category", "match": { "value": "docs" } }
  ]
}

topK

number

default:4

Number of results to return

fileUpload

boolean

Enable file uploads scoped to chat session

Usage Examples

Basic Local Setup

# Start Qdrant
docker run -p 6333:6333 -v $(pwd)/qdrant_storage:/qdrant/storage qdrant/qdrant

// In Flowise
Qdrant Server URL: http://localhost:6333
Qdrant Collection: my-documents
Vector Dimension: 1536
Similarity: Cosine
Embeddings: OpenAI Embeddings

Qdrant Cloud Production

// Cloud configuration
Qdrant Server URL: https://abc-example.cloud.qdrant.io
Credential: Qdrant API Key
Qdrant Collection: production-kb
Vector Dimension: 1536
Similarity: Cosine

Batch Upsert for Large Datasets

// Process large document sets efficiently
Document: PDF Folder Loader (1000+ files)
Qdrant Collection: large-corpus
Batch Size: 100
Vector Dimension: 1536
Embeddings: OpenAI Embeddings

With Advanced Filtering

// Complex filter conditions
{
  "qdrantFilter": {
    "must": [
      {
        "key": "metadata.category",
        "match": { "value": "technical" }
      },
      {
        "key": "metadata.year",
        "range": {
          "gte": 2023
        }
      }
    ],
    "should": [
      {
        "key": "metadata.priority",
        "match": { "value": "high" }
      }
    ]
  }
}

Custom Payload Keys

// Customize storage structure
Content Payload Key: text_content
Metadata Payload Key: doc_metadata
Vector Dimension: 1536

Qdrant Filter Syntax

Qdrant supports powerful filtering:

// Must conditions (AND)
{
  "must": [
    { "key": "category", "match": { "value": "docs" } },
    { "key": "language", "match": { "value": "en" } }
  ]
}

// Range filters
{
  "must": [
    {
      "key": "price",
      "range": {
        "gte": 10.0,
        "lt": 100.0
      }
    }
  ]
}

// Should conditions (OR)
{
  "should": [
    { "key": "tag", "match": { "value": "important" } },
    { "key": "tag", "match": { "value": "urgent" } }
  ]
}

// Must NOT conditions
{
  "must_not": [
    { "key": "status", "match": { "value": "archived" } }
  ]
}

// Nested conditions
{
  "must": [
    {
      "should": [
        { "key": "type", "match": { "value": "article" } },
        { "key": "type", "match": { "value": "blog" } }
      ]
    },
    { "key": "published", "match": { "value": true } }
  ]
}

// Array matching
{
  "must": [
    {
      "key": "tags",
      "match": {
        "any": ["AI", "ML", "NLP"]
      }
    }
  ]
}

Best Practices

Collection Design

Create separate collections per domain
Use descriptive collection names
Set appropriate vector dimensions
Choose right distance metric

Performance

Use batch upserts for large datasets
Enable indexing for frequent filters
Monitor collection size
Use appropriate batch sizes

Filtering

Index frequently filtered fields
Use payload schema for validation
Combine filters efficiently
Test filter performance

Scaling

Use Qdrant Cloud for production
Monitor memory usage
Implement sharding for huge datasets
Set up replicas for HA

Advanced Features

Payload Schema

Define schema for validation:

{
  "qdrantCollectionConfiguration": {
    "payload_schema": {
      "category": { "type": "keyword" },
      "year": { "type": "integer" },
      "rating": { "type": "float" }
    }
  }
}

Optimizer Configuration

{
  "qdrantCollectionConfiguration": {
    "optimizers_config": {
      "deleted_threshold": 0.2,
      "vacuum_min_vector_number": 1000,
      "default_segment_number": 5,
      "indexing_threshold": 20000
    }
  }
}

Quantization for Memory Efficiency

{
  "qdrantCollectionConfiguration": {
    "quantization_config": {
      "scalar": {
        "type": "int8",
        "quantile": 0.99
      }
    }
  }
}

Common Issues

Connection Failed

Can’t connect to Qdrant serverSolution:

Verify Qdrant is running: curl http://localhost:6333
Check URL format includes http/https
For cloud: verify API key is correct
Check firewall/network settings

Dimension Mismatch

Error: “Vector dimension does not match collection”Solution:

Set correct vector dimension parameter
Ensure embeddings match dimension
Delete and recreate collection if needed
Verify embedding model configuration

Collection Already Exists

Error when creating collectionSolution:

Collections are auto-created, this is normal
Use existing collection or delete old one
Check if dimension/metric matches

Slow Queries

Search is taking too longSolution:

Create indexes on filtered fields
Reduce topK value
Use more restrictive filters
Consider quantization
Check collection size and resources

Monitoring

Qdrant provides telemetry and monitoring:

# Health check
curl http://localhost:6333/healthz

# Collection info
curl http://localhost:6333/collections/my-collection

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

Metrics to monitor:

Vector count per collection
Query latency (p50, p95, p99)
Memory usage
Disk usage
Indexing status

Outputs

retriever

VectorStoreRetriever

Retriever interface with configured filters and topK

vectorStore

QdrantVectorStore

Direct vector store for custom operations

Overview

Language Models

Vector Stores

Document Loaders

Agents & Tools

Overview

Setup

Configuration

Required Parameters

Optional Parameters

Usage Examples

Basic Local Setup

Qdrant Cloud Production

Batch Upsert for Large Datasets

With Advanced Filtering

Custom Payload Keys

Qdrant Filter Syntax

Best Practices

Collection Design

Performance

Filtering

Scaling

Advanced Features

Payload Schema

Optimizer Configuration

Quantization for Memory Efficiency

Common Issues

Monitoring

Outputs

Build docs developers (and LLMs) love

Overview

Language Models

Vector Stores

Document Loaders

Agents & Tools

​Overview

​Setup

​Configuration

​Required Parameters

​Optional Parameters

​Usage Examples

​Basic Local Setup

​Qdrant Cloud Production

​Batch Upsert for Large Datasets

​With Advanced Filtering

​Custom Payload Keys

​Qdrant Filter Syntax

​Best Practices

Collection Design

Performance

Filtering

Scaling

​Advanced Features

​Payload Schema

​Optimizer Configuration

​Quantization for Memory Efficiency

​Common Issues

​Monitoring

​Outputs

​Related Resources

Build docs developers (and LLMs) love

Overview

Setup

Configuration

Required Parameters

Optional Parameters

Usage Examples

Basic Local Setup

Qdrant Cloud Production

Batch Upsert for Large Datasets

With Advanced Filtering

Custom Payload Keys

Qdrant Filter Syntax

Best Practices

Advanced Features

Payload Schema

Optimizer Configuration

Quantization for Memory Efficiency

Common Issues

Monitoring

Outputs

Related Resources