Vector Stores Reference

Reference for vector database providers supported by Mindwave's Brain system.

Overview

Mindwave supports multiple vector store providers for storing and searching embeddings. Vector stores are accessed through the Brain abstraction, which provides a consistent API across all providers.

Supported Providers

Pinecone

Managed vector database with excellent performance and scalability.

Configuration

bash

MINDWAVE_VECTORSTORE=pinecone
MINDWAVE_PINECONE_API_KEY=your-api-key
MINDWAVE_PINECONE_ENVIRONMENT=us-east1-gcp
MINDWAVE_PINECONE_INDEX=mindwave

Setup

Create account at pinecone.io
Create an index with matching dimensions (1536 for OpenAI ada-002)
Copy API key and environment from dashboard

Features

Managed service (no infrastructure)
High performance at scale
Built-in metadata filtering
Real-time updates
Auto-scaling

Best For

Production applications
High-volume search
Managed infrastructure
Teams without ops expertise

Qdrant

High-performance open-source vector database.

Configuration

bash

MINDWAVE_VECTORSTORE=qdrant
MINDWAVE_QDRANT_HOST=localhost
MINDWAVE_QDRANT_PORT=6333
MINDWAVE_QDRANT_API_KEY=  # Optional
MINDWAVE_QDRANT_COLLECTION=mindwave

Setup

Docker:

bash

docker run -p 6333:6333 qdrant/qdrant

Cloud: Available on Qdrant Cloud

Features

Open source
Self-hosted or cloud
Advanced filtering
Hybrid search support
Excellent performance

Best For

Self-hosted deployments
On-premise requirements
Cost optimization
Full control over infrastructure

Weaviate

Open-source vector database with ML models built-in.

Configuration

bash

MINDWAVE_VECTORSTORE=weaviate
MINDWAVE_WEAVIATE_URL=http://localhost:8080/v1
MINDWAVE_WEAVIATE_API_TOKEN=password
MINDWAVE_WEAVIATE_INDEX=items
MINDWAVE_WEAVIATE_ADDITIONAL_HEADERS='{}'

Setup

Docker:

bash

docker run -d \
  -p 8080:8080 \
  -e AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED=true \
  semitechnologies/weaviate:latest

Features

Open source
Built-in vectorization
GraphQL API
Multi-modal support
RESTful interface

Best For

Multi-modal applications
GraphQL integration
Built-in ML models
Complex data relationships

File Storage (Development Only)

Simple JSON file storage for development and testing.

Configuration

bash

MINDWAVE_VECTORSTORE=file
MINDWAVE_VECTORSTORE_PATH=storage/mindwave/vectorstore.json

Features

No dependencies
Easy debugging
Fast setup
Portable

Limitations

Not suitable for production
No concurrent access
No advanced filtering
Linear search (slow for large datasets)

Best For

Local development
Testing
Prototyping
Small datasets (< 1000 items)

Array Storage (Testing Only)

In-memory storage for unit tests.

Configuration

bash

MINDWAVE_VECTORSTORE=array

Features

Ephemeral (lost on restart)
Fast
No setup required
Perfect for testing

Best For

Unit tests
Integration tests
CI/CD pipelines
Ephemeral environments

Using Brain API

The Brain provides a consistent API across all vector stores:

Store Vectors

php

use Mindwave\Mindwave\Facades\Mindwave;

$brain = Mindwave::brain('documentation');

// Remember single item
$brain->remember('Laravel is a PHP framework', ['id' => 1, 'type' => 'doc']);

// Remember multiple items
$brain->rememberMany([
    ['text' => 'Vue.js is a framework', 'metadata' => ['id' => 2]],
    ['text' => 'Docker is a platform', 'metadata' => ['id' => 3]],
]);

Search Vectors

php

// Search by text (automatic embedding)
$results = $brain->recall('PHP web framework', limit: 5);

foreach ($results as $result) {
    echo $result->content;        // Original text
    echo $result->score;          // Similarity score (0-1)
    print_r($result->metadata);   // Metadata array
}

// Search by embedding
$embedding = Mindwave::embeddings()->embed('PHP framework');
$results = $brain->recallByEmbedding($embedding, limit: 5);

Filter by Metadata

php

// Store with metadata
$brain->remember('Laravel routing', ['category' => 'docs', 'version' => '10.x']);
$brain->remember('Vue components', ['category' => 'tutorial', 'version' => '3.x']);

// Filter during search
$results = $brain->recall('framework basics',
    limit: 5,
    filter: ['category' => 'docs']
);

Forget (Delete) Vectors

php

// Forget by metadata
$brain->forget(['id' => 1]);

// Clear all
$brain->forgetAll();

Performance Comparison

Feature	Pinecone	Qdrant	Weaviate	File	Array
Scale	Millions	Millions	Millions	< 10K	< 1K
Speed	Excellent	Excellent	Excellent	Slow	Fast
Setup	Easy	Medium	Medium	Easy	Easy
Cost	Paid	Free/Paid	Free/Paid	Free	Free
Filtering	Yes	Yes	Yes	Limited	Limited
Production	Yes	Yes	Yes	No	No

Configuration Examples

Development Setup

bash

# Simple file storage for local dev
MINDWAVE_VECTORSTORE=file
MINDWAVE_VECTORSTORE_PATH=storage/mindwave/vectors.json

Production Setup (Pinecone)

bash

MINDWAVE_VECTORSTORE=pinecone
MINDWAVE_PINECONE_API_KEY=${PINECONE_KEY}  # From secrets manager
MINDWAVE_PINECONE_ENVIRONMENT=us-east1-gcp
MINDWAVE_PINECONE_INDEX=production-vectors

Self-Hosted Setup (Qdrant)

bash

MINDWAVE_VECTORSTORE=qdrant
MINDWAVE_QDRANT_HOST=qdrant.internal
MINDWAVE_QDRANT_PORT=6333
MINDWAVE_QDRANT_API_KEY=${QDRANT_KEY}
MINDWAVE_QDRANT_COLLECTION=app-vectors

Migration Between Providers

Migrate data between vector stores:

php

use Mindwave\Mindwave\Facades\Mindwave;

// Source brain
$source = Mindwave::brain('old-brain');
$source->setDriver('file');

// Destination brain
$destination = Mindwave::brain('new-brain');
$destination->setDriver('pinecone');

// Get all vectors from source
$vectors = $source->getAllVectors(); // Depends on provider

// Store in destination
foreach ($vectors as $vector) {
    $destination->remember($vector->text, $vector->metadata, $vector->embedding);
}

Best Practices

Choose the Right Provider

Pinecone - Best for:

Production applications
Startups and small teams
Managed infrastructure preference

Qdrant - Best for:

Self-hosted requirements
Cost optimization
Full infrastructure control

Weaviate - Best for:

Multi-modal use cases
GraphQL integration
Built-in ML models

File/Array - Best for:

Development only
Testing only

Index Configuration

Configure dimensions to match your embedding model:

bash

# OpenAI ada-002 or 3-small
Dimensions: 1536

# OpenAI 3-large
Dimensions: 3072

Metadata Strategy

Use consistent metadata across your application:

php

$brain->remember($text, [
    'id' => $model->id,
    'type' => 'article',
    'category' => 'tech',
    'created_at' => now()->timestamp,
    'author_id' => $author->id,
]);

Batch Operations

Use batch operations for better performance:

php

// Bad: Individual inserts
foreach ($items as $item) {
    $brain->remember($item->text, ['id' => $item->id]);
}

// Good: Batch insert
$batch = $items->map(fn($item) => [
    'text' => $item->text,
    'metadata' => ['id' => $item->id]
])->toArray();

$brain->rememberMany($batch);

Troubleshooting

Connection Issues

Pinecone:

bash

# Test API key
curl -H "Api-Key: YOUR_KEY" https://controller.YOUR_ENV.pinecone.io/databases

Qdrant:

bash

# Test connection
curl http://localhost:6333/collections

Weaviate:

bash

# Test connection
curl http://localhost:8080/v1/.well-known/ready

Dimension Mismatch

Ensure vector dimensions match across:

Embedding model output
Vector store configuration
Query vectors

php

// Check embedding dimensions
$embedding = Mindwave::embeddings()->embed('test');
echo count($embedding); // Should match vector store config

Slow Queries

Optimize with:

Proper indexing
Metadata filtering
Limit result count
Batch operations

php

// Use filters to reduce search space
$results = $brain->recall($query,
    limit: 10,  // Reasonable limit
    filter: ['category' => 'relevant']  // Filter early
);

Brain API - Brain usage guide
Embeddings Reference - Embedding providers
RAG Overview - RAG architecture
Configuration Guide - Complete configuration

Vector Stores Reference ​

Overview ​

Supported Providers ​

Pinecone ​

Configuration ​

Setup ​

Features ​

Best For ​

Qdrant ​

Configuration ​

Setup ​

Features ​

Best For ​

Weaviate ​

Configuration ​

Setup ​

Features ​

Best For ​

File Storage (Development Only) ​

Configuration ​

Features ​

Limitations ​

Best For ​

Array Storage (Testing Only) ​

Configuration ​

Features ​

Best For ​

Using Brain API ​

Store Vectors ​

Search Vectors ​

Filter by Metadata ​

Forget (Delete) Vectors ​

Performance Comparison ​

Configuration Examples ​

Development Setup ​

Production Setup (Pinecone) ​

Self-Hosted Setup (Qdrant) ​

Migration Between Providers ​

Best Practices ​

Choose the Right Provider ​

Index Configuration ​

Metadata Strategy ​

Batch Operations ​

Troubleshooting ​

Connection Issues ​

Dimension Mismatch ​

Slow Queries ​

Related Documentation ​

Vector Stores Reference

Overview

Supported Providers

Pinecone

Configuration

Setup

Features

Best For

Qdrant

Configuration

Setup

Features

Best For

Weaviate

Configuration

Setup

Features

Best For

File Storage (Development Only)

Configuration

Features

Limitations

Best For

Array Storage (Testing Only)

Configuration

Features

Best For

Using Brain API

Store Vectors

Search Vectors

Filter by Metadata

Forget (Delete) Vectors

Performance Comparison

Configuration Examples

Development Setup

Production Setup (Pinecone)

Self-Hosted Setup (Qdrant)

Migration Between Providers

Best Practices

Choose the Right Provider

Index Configuration

Metadata Strategy

Batch Operations

Troubleshooting

Connection Issues

Dimension Mismatch

Slow Queries

Related Documentation