Embeddings

Embeddings are numerical representations of text that capture semantic meaning. They enable semantic search, similarity comparison, and are essential for vector databases and RAG applications.

What are Embeddings?

Embeddings convert text into high-dimensional vectors (arrays of numbers) where:

Similar meanings → Similar vectors
Different meanings → Distant vectors
Semantic relationships are preserved

For example, “king” and “queen” will have similar embeddings because they’re semantically related, even though they share no letters.

Why Use Embeddings?

Semantic Search

Find content by meaning, not just keywords

Similarity Comparison

Compare documents for similarity

RAG Applications

Provide relevant context to LLMs

Classification

Categorize content by semantic similarity

Available Embedding Models

n8n supports embeddings from multiple providers:

OpenAI Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsOpenAi Source Reference:

/home/daytona/workspace/source/packages/@n8n/nodes-langchain/nodes/embeddings/EmbeddingsOpenAI/EmbeddingsOpenAi.node.ts:78

Models:

text-embedding-3-small (1536 dimensions) - Default, best value
text-embedding-3-large (3072 dimensions) - Highest quality
text-embedding-ada-002 (1536 dimensions) - Legacy model

Configuration:

{
  "type": "@n8n/n8n-nodes-langchain.embeddingsOpenAi",
  "parameters": {
    "model": "text-embedding-3-small",
    "options": {
      "dimensions": 1536,
      "batchSize": 512,
      "stripNewLines": true
    }
  }
}

Options: From the source code:

options: {
  dimensions: 256 | 512 | 1024 | 1536 | 3072,
  batchSize: number,        // Max 2048
  stripNewLines: boolean,
  timeout: number,
  encodingFormat: 'float' | 'base64'
}

text-embedding-3-small offers the best balance of quality, speed, and cost for most use cases.

Cohere Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsCohere Source Reference:

/home/daytona/workspace/source/packages/@n8n/nodes-langchain/nodes/embeddings/EmbeddingsCohere/EmbeddingsCohere.node.ts:13

Models:

embed-english-v3.0 (1024 dimensions)
embed-english-v2.0 (4096 dimensions)
embed-english-light-v3.0 (384 dimensions)
embed-english-light-v2.0 (1024 dimensions)
embed-multilingual-v3.0 (1024 dimensions)
embed-multilingual-v2.0 (768 dimensions)
embed-multilingual-light-v3.0 (384 dimensions)

Configuration:

{
  "type": "@n8n/n8n-nodes-langchain.embeddingsCohere",
  "parameters": {
    "modelName": "embed-english-v3.0"
  }
}

Each Cohere model uses different dimensional density. Make sure your vector store matches the model’s dimensions.

Google Embeddings

Google Gemini Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsGoogleGemini Models:

text-embedding-004
embedding-001

Google Vertex AI Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsGoogleVertex Use when: You need Google Cloud integration

Azure OpenAI Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsAzureOpenAi Use when: You’re using Azure OpenAI Service Configuration:

{
  "type": "@n8n/n8n-nodes-langchain.embeddingsAzureOpenAi",
  "parameters": {
    "deploymentName": "text-embedding-ada-002"
  }
}

AWS Bedrock Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsAwsBedrock Models:

Amazon Titan Embeddings
Cohere Embed models via Bedrock

Ollama Embeddings (Local)

Node: @n8n/n8n-nodes-langchain.embeddingsOllama Use when: You need local/private embeddings Models:

nomic-embed-text
all-minilm
mxbai-embed-large
Any Ollama-compatible model

Configuration:

{
  "type": "@n8n/n8n-nodes-langchain.embeddingsOllama",
  "parameters": {
    "model": "nomic-embed-text",
    "baseUrl": "http://localhost:11434"
  }
}

Ollama is perfect for development or when you need to keep data private. No API costs!

Hugging Face Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsHuggingFaceInference Use when: You want open-source models via API Popular Models:

sentence-transformers/all-MiniLM-L6-v2 (384 dimensions)
sentence-transformers/all-mpnet-base-v2 (768 dimensions)
BAAI/bge-base-en-v1.5 (768 dimensions)

Mistral Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsMistralCloud Model: mistral-embed

Lemonade AI Embeddings

Node: @n8n/n8n-nodes-langchain.embeddingsLemonade Use when: Using Lemonade AI platform

Choosing an Embedding Model

Decision Matrix

Use Case	Recommended Model	Dimensions	Notes
General Purpose	OpenAI text-embedding-3-small	1536	Best value
High Quality	OpenAI text-embedding-3-large	3072	More expensive
Multilingual	Cohere embed-multilingual-v3.0	1024	Great for non-English
Budget-Conscious	Cohere embed-english-light-v3.0	384	Lower cost
Local/Private	Ollama nomic-embed-text	768	No API costs
Azure Integration	Azure OpenAI	1536	Native Azure
AWS Integration	Bedrock Titan	1536	Native AWS

Considerations

Quality Requirements

Higher dimensions generally mean better quality but higher costs

Language Support

Use multilingual models if you need non-English support

Cost

Consider API costs, especially at scale

Latency

Smaller models are faster to compute

Privacy

Use local models (Ollama) if data privacy is critical

Using Embeddings in Workflows

Basic RAG Pipeline

{
  "workflow": {
    "nodes": [
      {
        "name": "Load Documents",
        "type": "documentDefaultDataLoader"
      },
      {
        "name": "Split Text",
        "type": "textSplitterRecursiveCharacterTextSplitter",
        "parameters": {
          "chunkSize": 1000,
          "chunkOverlap": 200
        }
      },
      {
        "name": "Generate Embeddings",
        "type": "embeddingsOpenAi",
        "parameters": {
          "model": "text-embedding-3-small"
        }
      },
      {
        "name": "Store Vectors",
        "type": "vectorStorePinecone",
        "parameters": {
          "mode": "insert"
        }
      }
    ]
  }
}

Semantic Search

{
  "workflow": {
    "nodes": [
      {
        "name": "Embeddings",
        "type": "embeddingsOpenAi"
      },
      {
        "name": "Vector Store",
        "type": "vectorStorePinecone",
        "parameters": {
          "mode": "retrieve"
        }
      },
      {
        "name": "Retriever",
        "type": "retrieverVectorStore",
        "parameters": {
          "topK": 5
        }
      }
    ]
  }
}

Agent with Knowledge Base

{
  "workflow": {
    "nodes": [
      {
        "name": "Chat Trigger",
        "type": "chatTrigger"
      },
      {
        "name": "Embeddings",
        "type": "embeddingsOpenAi"
      },
      {
        "name": "Knowledge Base",
        "type": "vectorStorePinecone"
      },
      {
        "name": "Vector Store Tool",
        "type": "toolVectorStore",
        "parameters": {
          "name": "knowledge_base",
          "description": "Search the company knowledge base"
        }
      },
      {
        "name": "Agent",
        "type": "agent"
      }
    ]
  }
}

Embedding Configuration

Batch Size

From the OpenAI embeddings source code:

{
  displayName: 'Batch Size',
  name: 'batchSize',
  default: 512,
  typeOptions: { maxValue: 2048 },
  description: 'Maximum number of documents to send in each request'
}

Larger batch sizes improve throughput but may hit rate limits. Start with 512 and adjust based on your usage.

Dimensions (OpenAI only)

OpenAI’s text-embedding-3 models support configurable dimensions:

{
  displayName: 'Dimensions',
  name: 'dimensions',
  default: 1536,
  options: [256, 512, 1024, 1536, 3072],
  description: 'The number of dimensions the resulting output embeddings should have.'
}

Lower dimensions:

✅ Faster processing
✅ Lower storage costs
✅ Faster similarity search
❌ Lower quality

Higher dimensions:

✅ Better quality
✅ More accurate results
❌ Slower
❌ Higher costs

Strip New Lines

{
  displayName: 'Strip New Lines',
  name: 'stripNewLines',
  default: true,
  description: 'Whether to strip new lines from the input text'
}

Stripping new lines can improve embedding quality by normalizing text format.

Best Practices

Always Use the Same Model

Critical: Always use the same embedding model for both inserting and querying. Mixing models will produce incorrect results.

// ✅ Correct
Insert: embeddingsOpenAi (text-embedding-3-small)
Query:  embeddingsOpenAi (text-embedding-3-small)

// ❌ Wrong
Insert: embeddingsOpenAi (text-embedding-3-small)
Query:  embeddingsCohere (embed-english-v3.0)

Match Vector Store Dimensions

Ensure your vector store is configured with the correct dimensions:

// OpenAI text-embedding-3-small
Embedding dimensions: 1536
Vector store dimensions: 1536 ✅

// Cohere embed-english-v3.0
Embedding dimensions: 1024
Vector store dimensions: 1024 ✅

Optimize Chunk Size

Chunk your text before embedding:

{
  "type": "textSplitterRecursiveCharacterTextSplitter",
  "parameters": {
    "chunkSize": 1000,      // Characters per chunk
    "chunkOverlap": 200     // Overlap between chunks
  }
}

Guidelines:

Too small (< 200 chars): Lacks context
Too large (> 2000 chars): Dilutes specific information
Optimal: 500-1000 characters
Overlap: 10-20% of chunk size

Handle Rate Limits

{
  "options": {
    "batchSize": 512,       // Reduce if hitting rate limits
    "timeout": 60000        // Increase timeout for large batches
  }
}

Monitor Costs

OpenAI Pricing (as of 2024):

text-embedding-3-small: $0.02 / 1M tokens
text-embedding-3-large: $0.13 / 1M tokens
text-embedding-ada-002: $0.10 / 1M tokens

Cost Optimization:

Use smaller models when appropriate
Cache embeddings when possible
Use lower dimensions for text-embedding-3
Consider Ollama for development

Cache Embeddings

Avoid re-embedding the same content:

// Store document hash with embedding
{
  "metadata": {
    "contentHash": "abc123...",
    "embeddingModel": "text-embedding-3-small"
  }
}

Advanced Features

Custom Base URL

For OpenAI-compatible APIs:

{
  "options": {
    "baseURL": "https://api.custom-provider.com/v1"
  }
}

Custom Headers

Add custom HTTP headers:

{
  "options": {
    "customHeaders": {
      "X-Custom-Header": "value"
    }
  }
}

Timeout Configuration

{
  "options": {
    "timeout": 60000  // 60 seconds
  }
}

Troubleshooting

Dimension Mismatch Error

Error: “Expected vector of size X, got Y” Solution: Ensure vector store dimensions match embedding model:

// Check your embedding model dimensions
const modelDimensions = {
  'text-embedding-3-small': 1536,
  'text-embedding-3-large': 3072,
  'embed-english-v3.0': 1024,
  'embed-multilingual-v2.0': 768
};

// Configure vector store accordingly

Rate Limit Errors

Error: “Rate limit exceeded” Solutions:

Reduce batch size
Add delays between batches
Implement exponential backoff
Upgrade API tier

{
  "options": {
    "batchSize": 256  // Reduce from 512
  }
}

Poor Search Quality

Possible Causes:

Wrong embedding model for query
Poor chunk size
Missing context
Low-quality source data

Solutions:

Verify same model for insert and query
Adjust chunk size and overlap
Include surrounding context in chunks
Clean and preprocess source data

High Latency

Solutions:

Use faster models (smaller dimensions)
Increase batch size
Consider local models (Ollama)
Use edge locations
Implement caching

Comparison Table

Provider	Model	Dimensions	Cost (per 1M tokens)	Best For
OpenAI	text-embedding-3-small	1536	$0.02	General purpose
OpenAI	text-embedding-3-large	3072	$0.13	High quality
Cohere	embed-english-v3.0	1024	$0.10	English text
Cohere	embed-multilingual-v3.0	1024	$0.10	Multilingual
Cohere	embed-english-light-v3.0	384	$0.10	Budget
Ollama	nomic-embed-text	768	Free	Local/Private
Hugging Face	all-MiniLM-L6-v2	384	Free	Development

Next Steps

Vector Stores

Store your embeddings in vector databases

RAG Tutorial

Build a complete RAG application

Text Splitters

Learn about chunking strategies

Semantic Search

Implement semantic search

​Embeddings

​What are Embeddings?

​Why Use Embeddings?

Semantic Search

Similarity Comparison

RAG Applications

Classification

​Available Embedding Models

​OpenAI Embeddings

​Cohere Embeddings

​Google Embeddings

​Google Gemini Embeddings

​Google Vertex AI Embeddings

​Azure OpenAI Embeddings

​AWS Bedrock Embeddings

​Ollama Embeddings (Local)

​Hugging Face Embeddings

​Mistral Embeddings

​Lemonade AI Embeddings

​Choosing an Embedding Model

​Decision Matrix

​Considerations

​Using Embeddings in Workflows

​Basic RAG Pipeline

​Semantic Search

​Agent with Knowledge Base

​Embedding Configuration

​Batch Size

​Dimensions (OpenAI only)

​Strip New Lines

​Best Practices

​Always Use the Same Model

​Match Vector Store Dimensions

​Optimize Chunk Size

​Handle Rate Limits

​Monitor Costs

​Cache Embeddings

​Advanced Features

​Custom Base URL

​Custom Headers

​Timeout Configuration

​Troubleshooting

​Dimension Mismatch Error

​Rate Limit Errors

​Poor Search Quality

​High Latency

​Comparison Table

​Next Steps

Vector Stores

RAG Tutorial

Text Splitters

Semantic Search

Embeddings

What are Embeddings?

Why Use Embeddings?

Available Embedding Models

OpenAI Embeddings

Cohere Embeddings

Google Embeddings

Google Gemini Embeddings

Google Vertex AI Embeddings

Azure OpenAI Embeddings

AWS Bedrock Embeddings

Ollama Embeddings (Local)

Hugging Face Embeddings

Mistral Embeddings

Lemonade AI Embeddings

Choosing an Embedding Model

Decision Matrix

Considerations

Using Embeddings in Workflows

Basic RAG Pipeline

Semantic Search

Agent with Knowledge Base

Embedding Configuration

Batch Size

Dimensions (OpenAI only)

Strip New Lines

Best Practices

Always Use the Same Model

Match Vector Store Dimensions

Optimize Chunk Size

Handle Rate Limits

Monitor Costs

Cache Embeddings

Advanced Features

Custom Base URL

Custom Headers

Timeout Configuration

Troubleshooting

Dimension Mismatch Error

Rate Limit Errors

Poor Search Quality

High Latency

Comparison Table

Next Steps