Layered Memory Agent System

This project implements a Layered Memory architecture for LLM-based agents, separating memory into Semantic, Episodic, and Procedural layers, coordinated through a central Memory Manager and executed via an agent loop.

The goal is to move beyond a stateless chatbot and build an agent that remembers, reasons, uses tools, and decides when to stop.

---

High-Level Architecture

Core concepts:

• Semantic Memory
  Long-term factual knowledge (vector-based, Qdrant-backed).

• Episodic Memory
  Session-scoped conversation history and summaries.

• Procedural Memory
  Learned behaviors, tool usage patterns, and workflows.

• Memory Manager
  Controls what to read, what to write, and when.

• LLM Loop
  Reasons over memory, invokes tools, and decides continuation.

All essential flow diagrams are documented under docs and diagrams/.

---

Project Structure

.
├── docs and diagrams/
├── experiments/
├── qdrant_storage/
├── tests/
│
├── .env
├── layeredmemory.py
├── memory_core.py
├── streamlit_app.py
│
├── redis_ttl_test.py
├── working_memory.py

---

Environment Setup

Create a .env file:

OPENAI_API_KEY=your_key_here
QDRANT_URL=http://localhost:6333
QDRANT_COLLECTION=semantic_memory

---

Running the Project

streamlit run streamlit_app.py

---

Final Note

This repository is diagram-driven. Update diagrams before code changes.