Struct RecallItem 

pub struct RecallItem {
    pub memory_id: i64,
    pub name: String,
    pub namespace: String,
    pub memory_type: String,
    pub description: String,
    pub snippet: String,
    pub distance: f32,
    pub score: f32,
    pub source: String,
    pub graph_depth: Option<u32>,
}

Individual item returned by the recall query.

The memory_type field is serialised as "type" in JSON to maintain compatibility with external clients — the Rust name uses memory_type to avoid conflict with the reserved keyword.

Examples

use sqlite_graphrag::output::RecallItem;

let item = RecallItem {
    memory_id: 7,
    name: "nota-rust".into(),
    namespace: "global".into(),
    memory_type: "user".into(),
    description: "aprendizado de Rust".into(),
    snippet: "ownership e borrowing".into(),
    distance: 0.12,
    score: 0.88,
    source: "direct".into(),
    graph_depth: None,
};

let json = serde_json::to_string(&item).unwrap();
// Rust field `memory_type` appears as `"type"` in JSON.
assert!(json.contains("\"type\":\"user\""));
assert!(!json.contains("memory_type"));
assert!(json.contains("\"distance\":0.12"));

Fields

memory_id: i64

name: String

namespace: String

memory_type: String

description: String

snippet: String

distance: f32

score: f32

Cosine similarity in [0.0, 1.0] derived as 1.0 - distance and clamped to that interval. Always populated to satisfy the documented contract (M-A5 in v1.0.40); higher means more similar. For graph hits the value reflects the hop-derived distance proxy and should be interpreted alongside graph_depth rather than as a true cosine score.

source: String

graph_depth: Option<u32>

Number of graph hops between this match and the seed memories.

Set to None for direct vector matches (where distance is meaningful) and to Some(N) for traversal results, with N=0 when the depth could not be tracked precisely. Added in v1.0.23 to disambiguate graph results from the distance: 0.0 placeholder previously used for graph entries. Field is omitted from JSON output when None.

Implementations

impl RecallItem

pub fn score_from_distance(distance: f32) -> f32

Computes the similarity score from a vector distance, clamped to [0.0, 1.0]. Cosine distance returned by sqlite-vec lives in [0, 2] in theory but the embedder produces unit-norm vectors so the practical range is [0, 1]. Centralized so every constructor keeps the contract.

Trait Implementations

impl Clone for RecallItem

fn clone(&self) -> RecallItem

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Serialize for RecallItem

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.