Skip to content

Embedding Providers

Comprehensive guide to choosing and configuring embedding providers.

Overview

Embeddings are vector representations of text that enable semantic search. Nexus-Dev supports multiple providers to fit different needs:

Provider Best For Cost Privacy
OpenAI General use, easy setup $ Cloud
Ollama Privacy, offline use Free Local
Google Vertex AI GCP enterprises $$ Cloud
AWS Bedrock AWS enterprises $$ Cloud
Voyage AI Best RAG quality $$$ Cloud
Cohere Multilingual, reranking $$ Cloud

Provider Lock-in

Embeddings are NOT portable between providers. Changing providers requires re-indexing with nexus-reindex.

OpenAI (Default)

Recommended for: Getting started, general purpose.

Setup

export OPENAI_API_KEY="sk-..."

Configuration

{
  "embedding_provider": "openai",
  "embedding_model": "text-embedding-3-small"
}

Models

Model Dimensions Cost/1M tokens Quality
text-embedding-3-small 1536 $0.02 Good
text-embedding-3-large 3072 $0.13 Better
text-embedding-ada-002 1536 $0.10 Legacy

Start with Small

text-embedding-3-small offers excellent quality/cost ratio for most use cases.

Ollama (Local)

Recommended for: Privacy, offline use, cost savings.

Setup

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Start server
ollama serve

# Pull embedding model
ollama pull nomic-embed-text

Configuration

{
  "embedding_provider": "ollama",
  "embedding_model": "nomic-embed-text",
  "ollama_url": "http://localhost:11434"
}

Models

Model Dimensions Size Quality
nomic-embed-text 768 274MB Good
mxbai-embed-large 1024 669MB Better
all-minilm 384 45MB Fast

Performance

  • CPU: ~100-500ms per embedding
  • GPU: ~10-50ms per embedding
  • Batch: Process multiple files together for efficiency

Google Vertex AI

Recommended for: GCP enterprises, high scalability.

Setup

# Install extra
pipx install nexus-dev[google]

# Configure ADC
gcloud auth application-default login

Configuration

{
  "embedding_provider": "google",
  "embedding_model": "text-embedding-004",
  "google_project_id": "your-project-id",
  "google_location": "us-central1"
}

Models

Model Dimensions Notes
text-embedding-004 768 Latest, recommended
textembedding-gecko@003 768 Previous version

AWS Bedrock

Recommended for: AWS enterprises, existing AWS infrastructure.

Setup

# Install extra
pipx install nexus-dev[aws]

# Configure credentials
aws configure

Configuration

{
  "embedding_provider": "aws",
  "embedding_model": "amazon.titan-embed-text-v1",
  "aws_region": "us-east-1"
}

Models

Model Dimensions Notes
amazon.titan-embed-text-v1 1536 AWS native
cohere.embed-english-v3 1024 Via Bedrock

Voyage AI

Recommended for: Best-in-class RAG quality.

Setup

pipx install nexus-dev[voyage]

Configuration

{
  "embedding_provider": "voyage",
  "embedding_model": "voyage-large-2",
  "voyage_api_key": "your-key"
}

Models

Model Dimensions Best For
voyage-large-2 1024 General purpose
voyage-code-2 1536 Code search
voyage-lite-02-instruct 1024 Fast, cost-effective

For Code Search

voyage-code-2 is specifically trained for code and outperforms general models.

Cohere

Recommended for: Multilingual projects, reranking.

Setup

pipx install nexus-dev[cohere]

Configuration

{
  "embedding_provider": "cohere",
  "embedding_model": "embed-multilingual-v3.0",
  "cohere_api_key": "your-key"
}

Models

Model Dimensions Languages
embed-english-v3.0 1024 English
embed-multilingual-v3.0 1024 100+

Comparison

Quality Benchmarks

Based on MTEB (Massive Text Embedding Benchmark):

Provider Model MTEB Score
Voyage voyage-large-2 68.3
OpenAI text-embedding-3-large 64.6
Cohere embed-multilingual-v3 64.0
OpenAI text-embedding-3-small 62.3
Ollama nomic-embed-text 59.2

Cost Comparison

Approximate cost per 1M tokens indexed:

Provider Model Cost
Ollama nomic-embed-text $0 (local)
OpenAI text-embedding-3-small $0.02
OpenAI text-embedding-3-large $0.13
Cohere embed-english-v3 $0.10
Voyage voyage-large-2 $0.12

Choosing a Provider

Decision Tree

flowchart TD
    A[Start] --> B{Privacy Required?}
    B -->|Yes| C[Ollama]
    B -->|No| D{Enterprise Cloud?}
    D -->|GCP| E[Google Vertex AI]
    D -->|AWS| F[AWS Bedrock]
    D -->|No| G{Best Quality?}
    G -->|Yes, for code| H[Voyage voyage-code-2]
    G -->|Yes, general| I[Voyage voyage-large-2]
    G -->|Good enough| J[OpenAI text-embedding-3-small]

Quick Recommendations

Scenario Provider Model
Getting started OpenAI text-embedding-3-small
Privacy-sensitive Ollama nomic-embed-text
Maximum quality Voyage voyage-large-2
Code-focused Voyage voyage-code-2
Multilingual Cohere embed-multilingual-v3.0
GCP shop Google text-embedding-004
AWS shop AWS amazon.titan-embed-text-v1

Switching Providers

When changing providers:

  1. Update nexus_config.json
  2. Set required environment variables/credentials
  3. Re-index everything:
nexus-reindex

Full Re-index Required

You cannot mix embeddings from different providers. The entire database must be rebuilt.

See Also