Ollama Embeddings

The Ollama Embeddings node converts text into numerical vectors (embeddings) that capture meaning and context. Think of it as creating a “fingerprint” for text that allows AI to understand similarity and relationships between different pieces of content.

This is the foundation for building smart document search systems and AI knowledge bases that work completely locally on your machine.

How it works

The node takes your text content and uses a local Ollama model to convert it into a list of numbers (vectors) that represent the meaning of the text. These vectors can then be used to find similar content or build searchable databases.

graph LR
  Text[Your Text] --> Ollama[Ollama Model]
  Ollama --> Vectors[Number Vectors]
  Vectors --> Search[Searchable Database]
  style Ollama fill:#6d28d9,stroke:#fff,color:#fff

Setup guide

1. Install Ollama: Download and install Ollama on your computer from ollama.com.

2. Download Embedding Model: Run ollama pull nomic-embed-text to get a good embedding model.

3. Start Ollama Server: Run ollama serve to start the local server.

4. Configure Node: Set the Ollama URL and model name in your workflow.

Practical example: Building searchable knowledge base

Let’s convert website content into searchable vectors for a knowledge base.

Option 1: Basic Setup

• Server: http://localhost:11434
• Model: nomic-embed-text
• Content: The single text block you want to process.

Option 2: Batch Processing

• Goal: Process multiple texts at once for speed.
• Input: List of text strings.
• Batch Size: 5 items at a time.

Option 3: Custom Configuration

• Normalize: Enable to improve vector math consistency.
• Timeout: 30 seconds (give it more time for longer texts).

Why use local embeddings

Local Embeddings (Ollama)	Cloud Embeddings
Complete privacy - data never leaves your machine	Data sent to external servers
No ongoing costs	Pay per API call
Works offline	Requires internet connection
No rate limits	API rate limits apply
Full control over models	Limited model choices

Popular embedding models

Model	Best For	Size	Speed
nomic-embed-text	General purpose, good quality	~270MB	Fast
all-minilm	Lightweight, fast processing	~90MB	Very Fast
bge-large	High quality, detailed embeddings	~1.3GB	Slower

Start with nomic-embed-text

The nomic-embed-text model provides excellent quality for most use cases while being reasonably fast and not too large.

Real-world examples

Document similarity search

Create embeddings to find similar documents:

Input: "How to optimize website performance"
Output: [0.123, -0.456, 0.789, ...] (768 numbers)
Use: Find documents with similar vector patterns

Knowledge base creation

Convert company docs into searchable format:

Process: Company policies → Ollama Embeddings → Local Knowledge
Result: Smart search through all company documentation

Content recommendation

Find related articles or content:

Process: User reads article → Get embedding → Find similar embeddings
Result: "You might also like these articles"

Troubleshooting

• “Connection failed” errors: Make sure Ollama is running (ollama serve) and the URL is correct.
• “Model not found” errors: Download the model first with ollama pull nomic-embed-text.
• Slow processing: Try a smaller/faster model like all-minilm or reduce batch size.
• Out of memory errors: Use smaller batch sizes or try a lighter embedding model.