Embedding Providers API — OpenAI to Voyage

cohere

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/cohere"

Package cohere provides a Cohere embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the Cohere Embed API via the internal httpclient.

Registration

The provider registers as “cohere” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/cohere"

emb, err := embedding.New("cohere", config.ProviderConfig{
    APIKey: "...",
})

Models

Supported models and their default dimensions:

• embed-english-v3.0 (1024, default)
• embed-multilingual-v3.0 (1024)
• embed-english-light-v3.0 (384)
• embed-multilingual-light-v3.0 (384)
• embed-english-v2.0 (4096)

Configuration

ProviderConfig fields:

• APIKey — Cohere API key (required)
• Model — model name (default: “embed-english-v3.0”)
• BaseURL — API base URL (default: “https://api.cohere.com/v2”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality
• Options[“input_type”] — input type hint (default: “search_document”)

---

google

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/google"

Package google provides a Google AI embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the internal httpclient to call the Google AI Gemini embedding API.

Registration

The provider registers as “google” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/google"

emb, err := embedding.New("google", config.ProviderConfig{
    APIKey: "...",
})

Models

Supported models and their default dimensions:

• text-embedding-004 (768, default)
• embedding-001 (768)
• text-multilingual-embedding-002 (768)

Configuration

ProviderConfig fields:

• APIKey — Google AI API key (required)
• Model — model name (default: “text-embedding-004”)
• BaseURL — API base URL (default: “https://generativelanguage.googleapis.com/v1beta”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

inmemory

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/inmemory"

Package inmemory provides a deterministic hash-based Embedder for testing. It generates reproducible embeddings by hashing the input text with FNV-1a, making it suitable for unit tests and local development without external API calls. The resulting vectors are normalized to unit length.

Registration

The provider registers as “inmemory” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/inmemory"

emb, err := embedding.New("inmemory", config.ProviderConfig{})

Configuration

ProviderConfig fields:

• Options[“dimensions”] — vector size (default: 128)

---

jina

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/jina"

Package jina provides a Jina AI embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the Jina Embeddings API via the internal httpclient.

Registration

The provider registers as “jina” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/jina"

emb, err := embedding.New("jina", config.ProviderConfig{
    APIKey: "...",
})

Models

Supported models and their default dimensions:

• jina-embeddings-v2-base-en (768, default)
• jina-embeddings-v2-small-en (512)
• jina-embeddings-v2-base-de (768)
• jina-embeddings-v2-base-zh (768)
• jina-embeddings-v3 (1024)

Configuration

ProviderConfig fields:

• APIKey — Jina AI API key (required)
• Model — model name (default: “jina-embeddings-v2-base-en”)
• BaseURL — API base URL (default: “https://api.jina.ai/v1”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

mistral

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/mistral"

Package mistral provides an Embedder backed by the Mistral AI embeddings API. It implements the [embedding.Embedder] interface using Mistral’s embed endpoint via the internal httpclient.

Registration

The provider registers as “mistral” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/mistral"

emb, err := embedding.New("mistral", config.ProviderConfig{
    APIKey: "...",
    Model:  "mistral-embed",
})

Models

Supported models and their default dimensions:

• mistral-embed (1024, default)

Configuration

ProviderConfig fields:

• APIKey — Mistral AI API key (required)
• Model — model name (default: “mistral-embed”)
• BaseURL — API base URL (default: “https://api.mistral.ai/v1”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

ollama

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/ollama"

Package ollama provides an Ollama embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the Ollama REST API via the internal httpclient. Ollama enables running embedding models locally without external API calls.

Registration

The provider registers as “ollama” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/ollama"

emb, err := embedding.New("ollama", config.ProviderConfig{
    BaseURL: "http://localhost:11434",
})

Models

Supported models and their default dimensions:

• nomic-embed-text (768, default)
• mxbai-embed-large (1024)
• all-minilm (384)
• snowflake-arctic-embed (1024)

Configuration

ProviderConfig fields:

• Model — model name (default: “nomic-embed-text”)
• BaseURL — Ollama server URL (default: “http://localhost:11434”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

openai

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/openai"

Package openai provides an OpenAI embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the openai-go SDK.

Registration

The provider registers as “openai” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/openai"

emb, err := embedding.New("openai", config.ProviderConfig{
    APIKey: "sk-...",
})

Models

Supported models and their default dimensions:

• text-embedding-3-small (1536, default)
• text-embedding-3-large (3072)
• text-embedding-ada-002 (1536)

Configuration

ProviderConfig fields:

• APIKey — OpenAI API key (required)
• Model — model name (default: “text-embedding-3-small”)
• BaseURL — API base URL (default: “https://api.openai.com/v1”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

sentence_transformers

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/sentence_transformers"

Package sentencetransformers provides an Embedder backed by the HuggingFace Inference API for Sentence Transformers models. It implements the [embedding.Embedder] interface using the feature-extraction pipeline endpoint.

Registration

The provider registers as “sentence_transformers” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/sentence_transformers"

emb, err := embedding.New("sentence_transformers", config.ProviderConfig{
    APIKey: "hf_...",
    Model:  "sentence-transformers/all-MiniLM-L6-v2",
})

Models

Supported models and their default dimensions:

• sentence-transformers/all-MiniLM-L6-v2 (384, default)
• sentence-transformers/all-MiniLM-L12-v2 (384)
• sentence-transformers/all-mpnet-base-v2 (768)
• sentence-transformers/paraphrase-MiniLM-L6-v2 (384)
• BAAI/bge-small-en-v1.5 (384)
• BAAI/bge-base-en-v1.5 (768)
• BAAI/bge-large-en-v1.5 (1024)

Configuration

ProviderConfig fields:

• APIKey — HuggingFace API token (required)
• Model — model name (default: “sentence-transformers/all-MiniLM-L6-v2”)
• BaseURL — API base URL (default: “https://api-inference.huggingface.co”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality

---

voyage

import "github.com/lookatitude/beluga-ai/rag/embedding/providers/voyage"

Package voyage provides a Voyage AI embeddings provider for the Beluga AI framework. It implements the [embedding.Embedder] interface using the Voyage Embed API via the internal httpclient.

Registration

The provider registers as “voyage” in the embedding registry:

import _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/voyage"

emb, err := embedding.New("voyage", config.ProviderConfig{
    APIKey: "...",
})

Models

Supported models and their default dimensions:

• voyage-2 (1024, default)
• voyage-large-2 (1536)
• voyage-code-2 (1536)
• voyage-lite-02-instruct (1024)
• voyage-3 (1024)
• voyage-3-lite (512)

Configuration

ProviderConfig fields:

• APIKey — Voyage AI API key (required)
• Model — model name (default: “voyage-2”)
• BaseURL — API base URL (default: “https://api.voyageai.com/v1”)
• Timeout — request timeout
• Options[“dimensions”] — override output dimensionality
• Options[“input_type”] — input type hint (default: “document”)