Struct MemoryQuery 

pub struct MemoryQuery {
    pub text: Option<String>,
    pub category: Option<String>,
    pub tags: Vec<String>,
    pub agent: Option<String>,
    pub agent_prefix: Option<String>,
    pub limit: usize,
    pub memory_type: Option<MemoryType>,
    pub min_strength: Option<f64>,
    pub query_embedding: Option<Vec<f32>>,
    pub max_confidentiality: Option<Confidentiality>,
    pub reinforce: bool,
    pub exact_words: bool,
}

Query parameters for recalling memories.

limit controls the maximum number of results returned. A value of 0 means no limit (return all matching entries). This is the default.

Fields

text: Option<String>

category: Option<String>

tags: Vec<String>

agent: Option<String>

agent_prefix: Option<String>

Filter entries whose agent field starts with this prefix. Useful for cross-agent recall within a user namespace (e.g. "tg:123" matches "tg:123:assistant", "tg:123:researcher", etc.). Mutually exclusive with agent — if both are set, agent takes precedence.

limit: usize

Maximum number of results. 0 means unlimited.

memory_type: Option<MemoryType>

Filter by memory type.

min_strength: Option<f64>

Minimum strength threshold. Entries below this are excluded.

query_embedding: Option<Vec<f32>>

Optional query embedding for hybrid (BM25 + vector) retrieval. When present and entries have stored embeddings, cosine similarity is computed and fused with BM25 via Reciprocal Rank Fusion. Populated automatically by EmbeddingMemory::recall().

max_confidentiality: Option<Confidentiality>

When set, recall excludes entries with confidentiality above this level. None means no restriction (all levels returned).

reinforce: bool

Whether to reinforce the strength of returned entries on this read (Ebbinghaus reinforcement, +0.2 per access, capped at 1.0).

Defaults to true to preserve historical recall semantics. Set to false for a pure read — useful when surfacing strength to a UI, driving deterministic decay tests, or letting prune_weak_entries observe a freshly-stored low-strength entry without first promoting it above the prune threshold. last_accessed and access_count are still updated regardless.

exact_words: bool

Opt in to exact-word text matching instead of the default substring (word.contains(token)) semantics.

When true, InMemoryStore::recall short-circuits to entries whose lowercased content / keyword tokens exactly equal at least one query token, looked up via the in-memory inverted index built at store time. Estimated gain at N=10k entries: 12.69 ms → 1–3 ms text-query recall (Phase 8 in tasks/perf-audit-v2-2026-05-07.md).

Trade-off: queries whose tokens are prefixes / substrings of indexed words (“perf” matching “performance”) will no longer match. Default is false — substring semantics preserved; opt-in when callers know their queries are full words. The Postgres path ignores this flag (it doesn’t implement substring matching the same way).

Trait Implementations

impl Clone for MemoryQuery

fn clone(&self) -> MemoryQuery

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MemoryQuery

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for MemoryQuery

fn default() -> Self

Returns the “default value” for a type.