Embedding Providers

The RAG Starter Kit supports three embedding providers. The default is Google Gemini (free tier, 768D), which works out of the box with the database schema.

Provider Comparison

Provider	Model	Dimensions	Cost	Setup
Google Gemini	text-embedding-004	768	Free (1,500/day)	API key only
OpenAI	text-embedding-3-small	1536	$0.02/1M tokens	API key + migration
Ollama	nomic-embed-text	768	Free (local)	Local install

Google Gemini (Default)

No configuration needed beyond the API key. Produces 768D vectors matching the database schema.

.env

GOOGLE_GENERATIVE_AI_API_KEY=your-gemini-api-key
EMBEDDING_PROVIDER=google

Get a free API key at aistudio.google.com. Free tier includes 1,500 requests/day.

OpenAI

OpenAI embeddings produce 1536-dimensional vectors. You must run a database migration to change the vector column from 768 to 1536 dimensions before switching.

.env

EMBEDDING_PROVIDER=openai
EMBEDDING_MODEL=text-embedding-3-small
EMBEDDING_DIMENSIONS=1536
OPENAI_API_KEY=sk-...

Important: After changing the embedding provider, existing documents will have mismatched embeddings. Use the re-embed workspace function to regenerate all embeddings.

Ollama (Local)

Run embeddings locally with Ollama. The nomic-embed-text model produces 768D vectors compatible with the default schema.

.env

EMBEDDING_PROVIDER=ollama
EMBEDDING_MODEL=nomic-embed-text
OLLAMA_BASE_URL=http://localhost:11434

Install the model

ollama pull nomic-embed-text

Changing Providers

When you switch embedding providers, the system automatically detects the change and logs a warning. You should:

Update the .env variables
If dimensions changed, run a migration to update the vector column
Trigger re-embedding for all existing documents
Verify with the setup test endpoint: GET /api/setup/test

← Guides LLM Providers →