Crate ckg_embed 

Local embeddings via fastembed-rs (nomic-embed-text-v1.5, 768 dim).

Module layout:

• embedder — public API (Embedder, EMBED_DIM), lazy init, backend trait
• batching — byte-cap enforcement, batch chunking, output validation
• verify — ONNX SHA-256 hash verification (cache-poisoning gate)
• vec_math — cosine similarity helper

Re-exports

pub use vec_math::cosine;

Modules

vec_math

Cosine similarity helper.

Structs

Embedder

Lazy fastembed wrapper. Model is loaded on first embed() call.

Constants

EMBED_DIM

Expected embedding dimension for NomicEmbedTextV15.

Functions

has_cached_model

True if cache_dir holds an .onnx file at least min_size_bytes in size (recursive scan). Encapsulates the model-cache layout so callers (e.g. ckg setup) don’t reimplement the walk.