MemoRizz is an experimental Python library for building AI agents with explicit memory management, targeting developers and researchers in AI and LLM applications. It provides persistent memory, semantic search, tool integration, and persona systems, enabling context-aware agents that remember conversations and retrieve information efficiently.

How It Works

MemoRizz utilizes a modular architecture with a MemAgent as the core interface. It abstracts storage via a MemoryProvider interface, with a primary implementation leveraging MongoDB Atlas for persistent storage and vector search. OpenAI embeddings are used for semantic similarity searches across stored information, allowing agents to retrieve relevant data using natural language queries.

Quick Start & Requirements

• Install: pip install memorizz
• Prerequisites: Python 3.7+, MongoDB Atlas account (or local MongoDB with vector search), OpenAI API key.
• Setup: Requires configuring API keys and MongoDB connection URIs.
• Docs: Agent Memory Presentation, AIEWF Richmond's Talk

Highlighted Details

• Persistent memory storage and retrieval with semantic search via MongoDB Atlas Vector Search.
• Dynamic tool registration using decorators for LLM function calling.
• Persona system for defining and applying agent personalities and behaviors.
• Support for single and multi-agent architectures, including delegate patterns.

Maintenance & Community

This is an educational project, and contributions for learning purposes are welcome. The project is licensed under the MIT License.

Licensing & Compatibility

MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

MemoRizz is explicitly marked as EXPERIMENTAL and for EDUCATIONAL PURPOSES ONLY. It is not recommended for production environments or sensitive data due to potential bugs and lack of security audits. Active development may lead to breaking changes.