Skip to content

Memory Engine

QilbeeDB's memory engine provides bi-temporal memory storage for AI agents, tracking both when events occurred and when they were recorded.

Architecture

Agent Memory Interface
  ↓
Memory Types (Episodic | Semantic | Procedural | Factual)
  ↓
Bi-Temporal Storage (Event Time | Transaction Time)
  ↓
Consolidation Engine (Short-term → Long-term)

Memory Types

Episodic Memory

Personal experiences and events (conversations, observations).

Semantic Memory

General knowledge and facts.

Procedural Memory

How-to knowledge and procedures.

Factual Memory

Timestamped facts about entities.

Bi-Temporal Model

Every memory has two timestamps:

  • Event Time: When the event actually occurred
  • Transaction Time: When it was recorded in the database

This enables: - Historical queries - Corrections without data loss - Audit trail - Time-travel debugging

Consolidation

Memories automatically consolidate from short-term to long-term based on: - Relevance score - Access frequency - Time since creation - Relationships to other memories

Relevance Scoring

Each memory has a dynamic relevance score based on: 1. Recency: Recent memories score higher 2. Access Frequency: Frequently accessed memories score higher 3. Importance: Manually set importance level 4. Connections: Memories connected to many others score higher

Example Usage

from qilbeedb import QilbeeDB
from qilbeedb.memory import Episode

db = QilbeeDB("http://localhost:7474")
memory = db.agent_memory('assistant')

# Store conversation
episode = Episode.conversation(
    'assistant',
    'What is 2+2?',
    'The answer is 4'
)
memory.store_episode(episode)

# Recall recent episodes
recent = memory.recall(recency_hours=24, limit=10)

Next Steps