Struct IndexManager 

pub struct IndexManager {
    pub hnsw: HnswIndex,
    pub bm25: Bm25Index,
    pub bitmap: BitmapIndex,
    pub temporal: TemporalIndex,
    pub salience: SalienceIndex,
}

Owns all index types and provides unified indexing and hybrid search.

Fields

hnsw: HnswIndex

Vector similarity index.

bm25: Bm25Index

BM25 full-text index for keyword search.

bitmap: BitmapIndex

Tag and attribute bitmap index.

temporal: TemporalIndex

Timestamp range index.

salience: SalienceIndex

Importance score index.

Implementations

impl IndexManager

pub fn new(config: IndexManagerConfig) -> Self

Create a new index manager with the given configuration.

pub fn save(&self, dir: &Path) -> MenteResult<()>

Save all indexes to the given directory (bincode format).

pub fn load(dir: &Path) -> MenteResult<Self>

Load all indexes from the given directory. Tries .bin first, falls back to .json.

pub fn index_memory(&self, node: &MemoryNode)

Index a memory node across all indexes.

pub fn contains_vector(&self, id: MemoryId) -> bool

True when the memory’s vector is indexed and live.

pub fn vector_ids(&self) -> Vec<MemoryId>

All live vector indexed memory ids.

pub fn remove_vector_only(&self, id: MemoryId)

Tombstone a vector whose backing page no longer exists; used by open time reconciliation when a snapshot is newer than the last forget.

pub fn remove_memory(&self, id: MemoryId, node: &MemoryNode)

Remove a memory from all indexes.

pub fn hybrid_search( &self, query_embedding: &[f32], tags: Option<&[&str]>, time_range: Option<(Timestamp, Timestamp)>, k: usize, ) -> Vec<(MemoryId, f32)>

Hybrid search combining vector similarity, BM25 keyword matching, tag filtering, time range, and salience.

Strategy:

1. Vector search (HNSW) for top candidates
2. BM25 keyword search for top candidates
3. Merge via Reciprocal Rank Fusion (RRF)
4. Filter by tags and time range
5. Boost by salience and recency
6. Return top k results

pub fn hybrid_search_with_query( &self, query_embedding: &[f32], query_text: Option<&str>, tags: Option<&[&str]>, time_range: Option<(Timestamp, Timestamp)>, k: usize, ) -> Vec<(MemoryId, f32)>

Hybrid search with an optional text query for BM25 matching.

When query_text is provided, BM25 results are merged with vector results via RRF. When None, behaves like vector-only search.

pub fn hybrid_search_with_query_mode( &self, query_embedding: &[f32], query_text: Option<&str>, tags: Option<&[&str]>, tags_or: bool, time_range: Option<(Timestamp, Timestamp)>, k: usize, ) -> Vec<(MemoryId, f32)>

Hybrid search with configurable tag mode (AND vs OR).

Trait Implementations

impl Default for IndexManager

fn default() -> Self

Returns the “default value” for a type.