Module embeddings 

Vector embedding support for semantic recall in the Punch memory system.

Provides a BuiltInEmbedder (TF-IDF based, no external deps), an OpenAiEmbedder (calls the OpenAI embeddings API), and an EmbeddingStore backed by SQLite for persistence and similarity search.

Structs

BuiltInEmbedder

A TF-IDF vectorizer that works entirely offline with no external dependencies. Call BuiltInEmbedder::fit with a corpus to build the vocabulary, then use Embedder::embed to compute vectors.

Embedding

A vector embedding with its source text and metadata.

EmbeddingConfig

Configuration for the embedding engine.

EmbeddingStore

Persistent store for embeddings, backed by SQLite.

OpenAiEmbedder

An embedder that calls the OpenAI embeddings API.

Enums

EmbeddingProvider

Which backend to use for computing embeddings.

Traits

Embedder

Trait for computing vector embeddings from text.

Functions

bytes_to_vec

Deserialize little-endian bytes back to a Vec<f32>.

cosine_similarity

Compute the cosine similarity between two vectors.

top_k_similar

Return the top-k most similar embeddings to query_vec, sorted by descending similarity.

vec_to_bytes

Serialize a Vec<f32> to little-endian bytes.