Struct MemoryConfig 

pub struct MemoryConfig {

    pub mu1: f64,
    pub mu2: f64,
    pub alpha: f64,
    pub interleave_ratio: f64,
    pub replay_boost: f64,
    pub promote_threshold: f64,
    pub demote_threshold: f64,
    pub archive_threshold: f64,
    pub actr_decay: f64,
    pub context_weight: f64,
    pub importance_weight: f64,
    pub contradiction_penalty: f64,
    pub spacing_factor: f64,
    pub importance_floor: f64,
    pub consolidation_bonus: f64,
    pub forget_threshold: f64,
    pub reward_magnitude: f64,
    pub downscale_factor: f64,
    pub hebbian_enabled: bool,
    pub hebbian_threshold: i32,
    pub hebbian_decay: f64,
    pub stdp_enabled: bool,
    pub stdp_causal_threshold: f64,
    pub stdp_min_observations: i32,
    pub embedding: EmbeddingConfig,
    pub fts_weight: f64,
    pub embedding_weight: f64,
    pub actr_weight: f64,
    pub actr_sigmoid_center: f64,
    pub actr_sigmoid_scale: f64,
    pub entity_config: EntityConfig,
    pub entity_weight: f64,
    pub dedup_enabled: bool,
    pub dedup_threshold: f64,
    pub auto_extract_importance_cap: f64,
    pub recall_dedup_enabled: bool,
    pub recall_dedup_threshold: f64,
    pub temporal_weight: f64,
    pub hebbian_recall_weight: f64,
    pub adaptive_weights: bool,
    pub association: AssociationConfig,
    pub triple: TripleConfig,
    pub promotion: PromotionConfig,
}

All tunable parameters for the Engram memory system.

Default values come from neuroscience literature (ACT-R, Memory Chain Model, Ebbinghaus forgetting curve).

Fields

mu1: f64

Working memory decay rate (per day). Higher = faster decay.

mu2: f64

Core memory decay rate (per day). Higher = faster decay.

alpha: f64

Consolidation transfer rate (working → core per day)

interleave_ratio: f64

Fraction of archived memories replayed per cycle

replay_boost: f64

Core strength boost per replayed archived memory (base)

promote_threshold: f64

demote_threshold: f64

archive_threshold: f64

actr_decay: f64

Base-level activation decay parameter (d in t^-d)

context_weight: f64

Context spreading activation weight

importance_weight: f64

Importance weight in retrieval activation

contradiction_penalty: f64

Contradiction penalty in activation

spacing_factor: f64

Spacing effect multiplier

importance_floor: f64

Importance floor in stability

consolidation_bonus: f64

Consolidation bonus per consolidation count

forget_threshold: f64

Effective strength threshold for pruning

reward_magnitude: f64

Default reward magnitude

downscale_factor: f64

Global downscaling factor per consolidation cycle

hebbian_enabled: bool

Enable Hebbian link formation

hebbian_threshold: i32

Number of co-activations before link forms

hebbian_decay: f64

Link strength decay per consolidation cycle

stdp_enabled: bool

Enable temporal direction tracking

stdp_causal_threshold: f64

Forward/backward ratio threshold for causal inference

stdp_min_observations: i32

Minimum observations before STDP inference

embedding: EmbeddingConfig

Embedding provider configuration

fts_weight: f64

Weight for FTS exact matching in hybrid recall (0.0-1.0) Recommended: 0.15 for 15% FTS contribution

embedding_weight: f64

Weight for embedding similarity in recall scoring (0.0-1.0) Recommended: 0.60 for 60% semantic similarity contribution

actr_weight: f64

Weight for ACT-R activation in recall scoring (0.0-1.0) Recommended: 0.25 for 25% recency/frequency contribution Note: fts_weight + embedding_weight + actr_weight should sum to ~1.0

actr_sigmoid_center: f64

Sigmoid center for ACT-R activation normalization. Controls the “midpoint age” — memories with activation near this value get normalized to ~0.5. Default -5.5 ≈ 1-day-old single-access memory. Lower values shift the curve to favor older memories.

actr_sigmoid_scale: f64

Sigmoid scale for ACT-R activation normalization. Controls steepness: smaller = sharper transition, larger = gentler. Default 1.5 gives good discrimination across the 1min–30day range.

entity_config: EntityConfig

Entity extraction configuration

entity_weight: f64

Weight for entity matches in hybrid recall scoring (0.0-1.0)

dedup_enabled: bool

Enable dedup checking on write (default: true)

dedup_threshold: f64

Cosine similarity threshold for considering memories as duplicates (default: 0.95)

auto_extract_importance_cap: f64

Maximum importance for auto-extracted memories (default: 0.7). Prevents LLM extractor from assigning high importance to noise. Only affects memories stored via extraction pipeline, not manual add().

recall_dedup_enabled: bool

Enable dedup of recall results (default: true)

recall_dedup_threshold: f64

Cosine similarity threshold for recall result dedup (default: 0.85)

temporal_weight: f64

Weight for temporal channel in hybrid recall (0.0-1.0) Only meaningful when query has temporal indicators

hebbian_recall_weight: f64

Weight for Hebbian graph channel in hybrid recall (0.0-1.0)

adaptive_weights: bool

Enable query-type adaptive weight adjustment (default: true)

association: AssociationConfig

Write-time association discovery configuration

triple: TripleConfig

LLM triple extraction configuration

promotion: PromotionConfig

Knowledge promotion configuration

Implementations

impl MemoryConfig

pub fn chatbot() -> Self

Preset for conversational chatbots.

High replay, slow decay — optimized for long conversations.

pub fn task_agent() -> Self

Preset for short-lived task agents.

Fast decay, low replay — focus on recent task context.

pub fn personal_assistant() -> Self

Preset for long-term personal assistants.

Very slow core decay — remember preferences for months.

pub fn researcher() -> Self

Preset for research agents.

Minimal forgetting — everything might be relevant later.

Trait Implementations

impl Clone for MemoryConfig

fn clone(&self) -> MemoryConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MemoryConfig

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

Formats the value using the given formatter.

impl Default for MemoryConfig

fn default() -> Self

Literature-based defaults.

impl<'de> Deserialize<'de> for MemoryConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for MemoryConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.