Struct DecayEngine 

pub struct DecayEngine {
    pub config: DecayConfig,
}

Importance Decay Engine that runs as a background task.

Fields

config: DecayConfig

Implementations

impl DecayEngine

pub fn new(config: DecayConfig) -> Self

Create a new DecayEngine with the given configuration.

pub fn calculate_decay( &self, current_importance: f32, hours_elapsed: f64, memory_type: &MemoryType, access_count: u32, ) -> f32

Calculate decayed importance for a single memory.

Memory type determines base decay speed:

• Working: 3× faster (temporary by design)
• Episodic: 1× normal (events fade naturally)
• Semantic: 0.5× slower (knowledge persists)
• Procedural: 0.3× slower (skills are durable)

Usage pattern shields from decay (diminishing returns). Never-accessed memories fade 50% faster.

pub fn access_boost(current_importance: f32) -> f32

Calculate access boost for a memory that was just recalled. Scales with current importance — valuable memories get bigger boosts.

pub async fn apply_decay(&self, storage: &Arc<dyn VectorStorage>) -> DecayResult

Apply decay to all memories across all agent namespaces.

Iterates all namespaces prefixed with _dakera_agent_, loads memories, applies decay based on time since last access, and removes memories below the minimum importance threshold.

pub fn spawn( config: Arc<RwLock<DecayConfig>>, interval_secs: u64, storage: Arc<dyn VectorStorage>, metrics: Arc<BackgroundMetrics>, ) -> JoinHandle<()>

Spawn the decay engine as a background tokio task.

Takes a shared Arc<RwLock<DecayConfig>> so that config changes made at runtime via PUT /admin/decay/config take effect without a server restart. Each loop iteration re-reads the config before running, so strategy/half-life changes apply on the next cycle.