memory.rs — Memory Engine & Recall

The memory subsystem in Oxyde stores, ranks, and retrieves agent experiences. It is critical to long-term coherence, emotional relevance, and contextual prompting.

📦 Core Structs

pub struct Memory {
    pub text: String,
    pub timestamp: u64,
    pub tags: Vec<String>,
    pub emotion_vector: EmotionVector,
    pub importance: f32,
}

pub struct MemoryManager {
    pub memories: Vec<Memory>,
}

• Each memory has emotional metadata and a subjective importance score.

• Memories decay over time unless reinforced.

---

🔍 Recall Logic

The core method for memory retrieval is:

fn recall(&self, query: &str, top_k: usize) -> Vec<Memory>

It uses:

• TF-IDF-style string similarity

• Emotion vector alignment (valence-aware)

• Recency and priority weighting

Recall is weighted so emotionally significant or frequently accessed memories are more likely to surface.

---

🧠 Writing New Memories

You can store memories manually:

agent.memory.store("He gave me the crystal.", Some("important"));

Or let them be auto-written after every LLM response via Agent::receive_input.

---

🔁 Memory Decay & Compression (WIP)

Future roadmap includes:

• Memory decay based on age & access

• Compression of low-priority memories

• Persistent DB-backed memory

---

💡 Use Cases

• Agent remembers specific player actions

• NPC personalities evolve over time

• Behavior changes based on past relationships