AI Agent Memory System
A structured long-term memory subsystem for AI agents — persistent user memory, semantic retrieval, and full memory-lifecycle management.
- Python
- FastAPI
- Pydantic
- SQLite
- Hugging Face
- NumPy
- n8n
- Docker
- ngrok
Project overview
AI Agent Memory System is a structured, long-term memory architecture built for conversational AI agents, designed so an agent can retain and recall meaningful information about a user across multiple conversations instead of starting fresh every time. Rather than treating memory as raw conversation text, the system models each memory as a structured record — with fields like category, key, value, importance, confidence, and embedding — enabling reliable creation, updating, deletion, semantic retrieval, and expiration of memories.
Preview
Preview image unavailable.
Key features
- RESTful FastAPI backend with SQLite persistence.
- Structured memory records with confidence and importance scoring.
- Embedding-based semantic retrieval.
- Short-term vs long-term memory separation.
- n8n workflow with WhatsApp interface and ngrok tunneling.
- Dockerized deployment.
Technologies
- Python
- FastAPI
- Pydantic
- SQLite
- Hugging Face
- NumPy
- n8n
- Docker
- ngrok
What I learned
I learned how to give AI agents persistent memory as structured, queryable records instead of raw chat history — including embedding-based semantic retrieval, relevance scoring, and safe lifecycle management.