Module embedder 

Local embedding generation backed by fastembed. fastembed wrapper and per-process embedding cache.

Owns the in-process TextEmbedding model and exposes batch encode/query helpers used by remember, recall, and related commands.

Functions

controlled_batch_count

Returns the number of batches that embed_passages_controlled would produce for the given token_counts slice without running inference.

embed_passage

Embeds a single passage using the passage: prefix required by E5 models.

embed_passages_batch

Embeds multiple passages in a single ONNX batch call.

embed_passages_controlled

Embeds passages grouped into token-budget-aware batches to avoid OOM on variable-length inputs.

embed_passages_serial

Embed multiple passages one-by-one (serial ONNX inference).

embed_query

Embeds a search query using the query: prefix required by E5 models.

f32_to_bytes

Convert &[f32] to &[u8] for sqlite-vec storage.

get_embedder

Returns the process-wide singleton embedder, initializing it on first call. Subsequent calls return the cached instance regardless of models_dir.