embedder

Documentation

Overview

Package embedder provides implementations of the rag.Embedder interface for converting text into dense vector embeddings. Each implementation talks to a different backend (OpenAI, Azure OpenAI, Ollama) via plain HTTP — no additional SDK dependencies are required.

Index

• func DefaultDimensions(backend string) int
• func NewFromEnv() (rag.Embedder, error)
• func ValidateForRAG(log *slog.Logger) error
• type OllamaConfig
• type OllamaEmbedder
  • func NewOllamaEmbedder(cfg *OllamaConfig) *OllamaEmbedder
  • func (e *OllamaEmbedder) Embed(ctx context.Context, texts []string) ([][]float32, error)
• type OpenAIConfig
• type OpenAIEmbedder
  • func NewOpenAIEmbedder(cfg *OpenAIConfig) *OpenAIEmbedder
  • func (e *OpenAIEmbedder) Embed(ctx context.Context, texts []string) ([][]float32, error)

Functions

func DefaultDimensions

func DefaultDimensions(backend string) int

DefaultDimensions returns the correct default embedding vector size for the given backend name. Callers that need to pre-configure a vector store (e.g. Qdrant collection creation) should use this rather than hardcoding a value. EMBEDDING_DIMENSIONS always takes precedence when set.

func NewFromEnv

func NewFromEnv() (rag.Embedder, error)

NewFromEnv constructs a rag.Embedder using cascading defaults that inherit from the chat provider configuration when embedding-specific overrides are not set.

Resolution order:

1. EMBEDDING_PROVIDER — if unset, inherits MODEL_PROVIDER (default: ollama)
2. Per-backend credentials are inherited from the chat provider's env vars
3. EMBEDDING_MODEL — overrides the default model for the resolved backend
4. EMBEDDING_API_KEY — overrides the inherited API key
5. EMBEDDING_ENDPOINT — overrides the inherited endpoint
6. EMBEDDING_DIMENSIONS — overrides the default dimensions (ollama: 768, openai/azure: 1536)

func ValidateForRAG

func ValidateForRAG(log *slog.Logger) error

ValidateForRAG checks that the embedder configuration is safe to use when QDRANT_HOST is set. It returns an error if the configuration is clearly broken (e.g. azure embedder with no API key), and logs a warning if EMBEDDING_MODEL looks like a chat model rather than an embedding model.

This is a pre-flight check — call it before constructing the embedder or the Qdrant store so operators get a clear error at startup rather than a cryptic failure during the first embed call.

Types

type OllamaConfig

type OllamaConfig struct {
	// Host is the Ollama server base URL (e.g. "http://localhost:11434").
	Host string
	// Model is the embedding model name (e.g. "nomic-embed-text").
	Model string
}

OllamaConfig holds the settings for constructing an OllamaEmbedder.

type OllamaEmbedder

type OllamaEmbedder struct {
	// contains filtered or unexported fields
}

OllamaEmbedder implements rag.Embedder using the Ollama /api/embed endpoint. It is safe for concurrent use. No API key is required — Ollama runs locally.

func NewOllamaEmbedder

func NewOllamaEmbedder(cfg *OllamaConfig) *OllamaEmbedder

NewOllamaEmbedder constructs an OllamaEmbedder from the given config.

func (*OllamaEmbedder) Embed

func (e *OllamaEmbedder) Embed(ctx context.Context, texts []string) ([][]float32, error)

Embed converts a batch of texts into their corresponding embeddings. The returned slice is parallel to the input slice.

type OpenAIConfig

type OpenAIConfig struct {
	// BaseURL is the API base URL. For OpenAI: "https://api.openai.com/v1".
	// For Azure: "https://<resource>.openai.azure.com/openai".
	BaseURL string
	// APIKey is the authentication key.
	APIKey string
	// Model is the embedding model name (e.g. "text-embedding-3-small").
	Model string
	// Dimensions is the desired vector length (0 = model default).
	Dimensions int
	// Azure enables Azure OpenAI mode (api-key header + api-version param).
	Azure bool
	// APIVersion is the Azure OpenAI API version (e.g. "2025-04-01-preview").
	// Ignored when Azure is false.
	APIVersion string
}

OpenAIConfig holds the settings for constructing an OpenAIEmbedder.

type OpenAIEmbedder

type OpenAIEmbedder struct {
	// contains filtered or unexported fields
}

OpenAIEmbedder implements rag.Embedder using the OpenAI (or Azure OpenAI) embeddings REST API. It is safe for concurrent use.

func NewOpenAIEmbedder

func NewOpenAIEmbedder(cfg *OpenAIConfig) *OpenAIEmbedder

NewOpenAIEmbedder constructs an OpenAIEmbedder from the given config.

func (*OpenAIEmbedder) Embed

func (e *OpenAIEmbedder) Embed(ctx context.Context, texts []string) ([][]float32, error)

Embed converts a batch of texts into their corresponding embeddings. The returned slice is parallel to the input slice.