Struct HybridStore 

pub struct HybridStore { /* private fields */ }

Executes hybrid graph + vector queries.

Implementations

impl HybridStore

pub fn query( &self, q: HybridQuery<'_>, col: &Collection, ) -> Result<Vec<HybridResult>>

Run a hybrid graph + vector query.

The algorithm proceeds in three stages:

1. Graph traversal — walks the memory graph from q.anchor_node up to q.graph_depth hops, recording the maximum edge weight seen for each reachable node.
2. Vector search — retrieves the top q.top_k × 20 approximate nearest neighbours from the named collection.
3. Score blending — for each candidate, computes rank = q.alpha × vec_similarity + (1 − q.alpha) × graph_weight, then returns the top q.top_k results sorted by rank descending.

pub fn tri_modal_query( &self, q: &TriModalQuery, col: &Collection, ) -> Result<Vec<TriModalResult>>

Run a tri-modal graph + vector + FTS query.

The algorithm runs three searches and blends results:

1. Vector ANN — retrieves top_k × 20 approximate nearest neighbours.
2. Graph traversal — walks the memory graph from q.anchor_node up to q.graph_depth hops, recording the maximum edge weight per reachable node.
3. FTS keyword search — BM25 full-text search over the collection’s FTS index.

Each component is min-max normalised to [0, 1] within its own result set, then blended as final_score = alpha × vec_score + beta × graph_weight + gamma × fts_score.

The weights must satisfy alpha + beta + gamma ≈ 1.0 (tolerance ±0.01).