Numpy-based vector store for chunk embeddings. Stores vectors in a .npz file with two arrays: - chunk_ids: 1D array of chunk ID strings -

BM25-based search engine for RLM chunks. Features: - French/English tokenization - Fast BM25S scoring - Returns ranked results w

Embedding provider using FastEmbed (paraphrase-multilingual-MiniLM-L12-v2). 384 dimensions, higher accuracy, slightly slower. ONNX-based.

Embedding provider using Model2Vec (minishlab/potion-multilingual-128M). 256 dimensions, very fast inference, good multilingual support.

Abstract base class for embedding providers.

Records VectorStore calls so tests can assert sync happened.

Test _normalize_bm25_scores function.

Tests for backward compatibility with pre-Phase 9 chunks.

Tests for chunk_type storage in YAML frontmatter and index.

Tests for chunk_type parameter validation.

Tests for date parsing and range checking helpers.

Edge case tests.

Tests for the entity matching helper.

Tests for the core entity extraction function.

Tests for fuzzy grep combined with temporal filtering.

Test that everything works when semantic deps are absent.

Tests for entity filtering in grep().

Basic fuzzy grep functionality tests.

Edge cases and error handling for fuzzy grep.

Test fuzzy grep with project/domain filters.

Integration tests for fuzzy grep via grep() dispatcher.

Tests for temporal filtering in grep.

Test hybrid BM25 + semantic fusion logic.

Test that YAML metadata improves BM25 search ranking.

Tests for accent normalization.

Real-world fuzzy search scenarios.

Tests for entity filtering in search().

Tests for temporal filtering in BM25 search.

Tests for French/English tokenization.

Test VectorStore add/search/save/load/remove operations.