Zum Inhalt

embedder

module wittgenstein_pgvector_qsql.embedder

Embedder implementations for Ollama and OpenAI, implementing LangChain's Embeddings contract.

Both classes expose .embed(text) for single-text use (gate, direct callers) plus .embed_query() / .embed_documents() to satisfy the LangChain contract so they can be passed directly to PGVector's embeddings parameter. They are now thin wrappers over wittgenstein_embeddings — the generic provider abstraction shared with wittgenstein_kg_client — kept under their original names/signatures so existing callers (gate/embeddings.py, training/service.py) don't need to change.

build_embedder(config) dispatches by provider via duck-typing: any object that has the required attributes works — no inheritance needed.

Classes

  • EmbedderConfig — Structural protocol for embedder configuration.

  • OllamaEmbedder — Sync HTTP client for Ollama /api/embeddings + LangChain contract.

  • OpenAIEmbedder — Sync OpenAI embeddings client + LangChain contract.

Functions

  • build_embedder — Factory that returns the embedder for the active provider.

wittgenstein_pgvector_qsql.embedder.EmbedderConfig

class EmbedderConfig()

Bases : Protocol

Structural protocol for embedder configuration.

Any object exposing these attributes works with build_embedder — including an app's Settings, the core's config, or a plain dataclass.

wittgenstein_pgvector_qsql.embedder.OllamaEmbedder

class OllamaEmbedder(host: str, model: str, timeout_seconds: float = 60.0)

Bases : LangChainEmbeddingAdapter

Sync HTTP client for Ollama /api/embeddings + LangChain contract.

Logic now lives in wittgenstein_embeddings.OllamaEmbeddingClient — this class is a backward-compat name/shape for existing callers.

Methods

wittgenstein_pgvector_qsql.embedder.OllamaEmbedder.embed

method OllamaEmbedder.embed(text: str) → list[float]

wittgenstein_pgvector_qsql.embedder.OpenAIEmbedder

class OpenAIEmbedder(api_key: str, model: str = 'text-embedding-3-small')

Bases : LangChainEmbeddingAdapter

Sync OpenAI embeddings client + LangChain contract.

Logic now lives in wittgenstein_embeddings.OpenAIEmbeddingClient — this class is a backward-compat name/shape for existing callers.

Methods

wittgenstein_pgvector_qsql.embedder.OpenAIEmbedder.embed

method OpenAIEmbedder.embed(text: str) → list[float]

wittgenstein_pgvector_qsql.embedder.build_embedder

build_embedder(config: Any) → Embeddings

Factory that returns the embedder for the active provider.

Accepts any object with the `EmbedderConfig` attribute set (duck-typed)

an app's Settings, the lib's QSQLSettings, or a plain dataclass all work.

Parameters

  • config : Any — any object with embedding_provider and the provider-specific fields listed in EmbedderConfig.

Raises

  • ValueError — unknown embedding_provider or a required API key is missing.