Module chunks 

Searchable chunks store backed by SQLite FTS5.

Port reference: Aonyx RAG rag_system/utils/bm25_store.py + utils/hybrid_search.py.

V1 scope (this file)

• Chunk = a piece of text + project + source + timestamp + free-form metadata.
• SQLite FTS5 virtual table provides BM25-ranked full-text search out of the box, with a unicode61 remove_diacritics 2 tokenizer that survives accents.
• search_bm25(project?, query, k) returns the top-k chunks ordered by relevance, with positive score = -bm25(...) so larger = better.

V1.1 (deferred)

• Local embeddings via fastembed-rs (ONNX, ~30 MB model).
• HNSW index for vector ANN search.
• RRF fusion with k = 60 combining BM25 + vectors.
• Exponential temporal boost on recent chunks.

The trait signature already accepts a mode field so V1.1 can extend it without breaking callers.

Structs

Chunk

A piece of indexable text.

ScoredChunk

A search hit: a chunk and its score (larger = more relevant).

SqliteChunksStore

SQLite-backed ChunksStore using FTS5 for BM25 ranking.

Traits

ChunksStore

Async chunks store.

Type Aliases

ChunkId

Stable identifier for a Chunk.