Struct IndexSet 

pub struct IndexSet { /* private fields */ }

Implementations

impl IndexSet

pub fn new( build_params: impl Fn() -> Result<HnswParams, CoreError>, ) -> Result<Self, CoreError>

pub fn insert_atomic( &mut self, id: u64, memory_id: &str, embedding: &ThreeFieldEmbedding, rng_value: f64, ) -> Result<(), CoreError>

pub fn resolve_node_id(&self, node_id: u64) -> Option<&str>

pub fn rebuild_id_map(&mut self, entries: impl Iterator<Item = (u64, String)>)

pub fn search( &self, query: &[f32], top_k: usize, ) -> Result<Vec<(u64, f32)>, HnswError>

pub fn find_duplicate( &self, embedding: &ThreeFieldEmbedding, threshold: f32, ) -> Result<Option<(u64, f32)>, HnswError>

Find the best existing near-duplicate of embedding under the ALL-THREE policy: a candidate qualifies only when its context, action AND result similarities each exceed threshold. Returns the qualifying node with the highest minimum-field similarity, or None. Uses only per-graph HNSW search (top_k=K) intersected across the three fields — with top_k=1 the three graphs return different top nodes and the all-three condition would almost never hold.

pub fn delete(&mut self, id: u64) -> Result<(), HnswError>

pub fn contains(&self, id: u64) -> bool

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

pub fn serialize<W: Write>(&self, writer: &mut W) -> Result<()>

pub fn deserialize<R: Read>(reader: &mut R) -> Result<Self, HnswError>