I've been building an experimental memory system for AI agents called Memoria. It is a lightweight, encrypted backend that ranks memories by "emotional intensity".

Core features - ~180 MB footprint, zero external services or deps - End-to-end encryption using SQLCipher - Vector search (cosine similarity, NumPy only) - Emotion-based reinforcement and forgetting - Async Python API with cache layer - CI: pytest, ruff, mypy, coverage via Codecov

No Pinecone, no Redis, no cloud infra.

GitHub repo: https://github.com/Ingver1/memoria

I'd love feedback, ideas, or criticism. Contributors welcome!