Embedding Providers

textrawl supports three embedding providers: OpenAI (cloud), Google AI (cloud), and Ollama (local).

Provider Comparison

Feature	OpenAI	Google AI	Ollama
Location	Cloud	Cloud	Local
Privacy	Data sent to API	Data sent to API	Data stays local
Dimensions	1536	768	1024 or 768
Model	text-embedding-3-small	text-embedding-004	nomic-embed-text / v2-moe
Cost	Pay per token	Free tier available	Free
Speed	~100ms/request	~80ms/request	~50ms/request
Setup	API key	API key	Docker container

OpenAI Setup

# .env
EMBEDDING_PROVIDER=openai
OPENAI_API_KEY=sk-proj-...

Use the database schema for 1536 dimensions:

# Run in Supabase SQL Editor
scripts/setup-db.sql

Google AI Setup

# .env
EMBEDDING_PROVIDER=google
GOOGLE_AI_API_KEY=your-google-ai-api-key
GOOGLE_EMBEDDING_MODEL=text-embedding-004

Use the database schema for 768 dimensions:

# Run in Supabase SQL Editor
scripts/setup-db-google.sql

Ollama Setup

# Start Ollama
docker run -d --name ollama -v ollama:/root/.ollama -p 11434:11434 ollama/ollama
 
# Pull the model
docker exec -it ollama ollama pull nomic-embed-text

Configure environment:

# .env
EMBEDDING_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=nomic-embed-text

Use the database schema for 1024 dimensions:

# Run in Supabase SQL Editor
scripts/setup-db-ollama.sql

Switching Providers

Warning: You cannot mix providers. Switching requires re-embedding all documents.

To switch:

Export your documents (or keep source files)
Drop existing tables
Run new schema (different dimensions)
Re-upload all documents

The table below compares published MTEB benchmarks for OpenAI and Ollama's nomic-embed-text. Google AI's text-embedding-004 performs comparably on standard retrieval benchmarks but is not included due to differing evaluation methodology.

Benchmark	OpenAI	nomic-embed-text
MTEB Average	62.3	59.4
Retrieval	54.9	52.8
Clustering	45.0	44.2

OpenAI has a slight edge in benchmarks, but Ollama offers privacy and no API costs. Google AI offers a good balance with a free tier and competitive retrieval quality.

Recommendations

Choose OpenAI if:

You need maximum quality
Data privacy isn't critical
You prefer managed infrastructure

Choose Google AI if:

You want cloud convenience with a free tier
768-dimension embeddings are sufficient
You already use Google Cloud

Choose Ollama if:

Data must stay local
You want to avoid API costs
You have GPU resources
You're running fully self-hosted