Module embedder 

Local text embedder — generates vector embeddings for semantic search.

Ported from v2’s local_embedder.py (313 lines). Provides a trait-based embedder abstraction with two implementations:

• HttpEmbedder: Calls llama-server’s /v1/embeddings endpoint (preferred, no model download needed — uses the already-running llama-server)
• StubEmbedder: Hash-based pseudo-embeddings for testing/fallback

Future: An ort (ONNX Runtime) embedder can implement the same trait for fully local embeddings without a server dependency.

Environment Variables

Variable	Default	Description
WM_EMBEDDER_ENDPOINT	—	llama-server HTTP URL (e.g. http://localhost:8080)
WM_EMBEDDER_MODEL	local	Model name for the embeddings API
WM_EMBEDDER_DIM	384	Expected embedding dimensionality
WM_EMBEDDER_TIMEOUT_MS	30000	Request timeout in milliseconds

Structs

EmbedderConfig

Configuration for the HTTP-based embedder.

HttpEmbedder

HTTP-based embedder using llama-server’s /v1/embeddings endpoint.

StubEmbedder

Stub embedder — hash-based pseudo-embeddings for testing/fallback.

Traits

Embedder

Trait for text embedding providers.

Functions

create_embedder

Create an embedder from environment configuration, with fallback chain.

is_endpoint_safe

Validate an embedder endpoint URL for SSRF safety.