Memary: The Open Source Memory Layer for Autonomous Agents

Introduction

Memary is an innovative open-source project that provides a sophisticated memory layer for autonomous agents. Designed to emulate human memory, Memary equips AI agents with enhanced reasoning, learning, and self-improvement capabilities. By integrating advanced concepts like knowledge graphs and memory modules, Memary offers a robust framework for managing agent memories, making them more effective and intelligent.

Installation

Getting started with Memary is straightforward. To install via pip, ensure you are running Python version <= 3.11.9, then execute:

pip install memary

For a local installation:

1. Create a virtual environment with Python version <= 3.11.9.
2. Install dependencies:

   pip install -r requirements.txt
   

Memary supports local models via Ollama (e.g., Llama 3, LLaVA) and also integrates with OpenAI models. You'll need to set up a .env file with necessary API keys for services like OpenAI, Perplexity, Google Maps, and database connections (FalkorDB or Neo4j).

Examples

Memary provides a ChatAgent class for easy integration. Basic Usage:

from memary.agent.chat_agent import ChatAgent

system_persona_txt = "data/system_persona.txt"
user_persona_txt = "data/user_persona.txt"
past_chat_json = "data/past_chat.json"
memory_stream_json = "data/memory_stream.json"
entity_knowledge_store_json = "data/entity_knowledge_store.json"
chat_agent = ChatAgent(
    "Personal Agent",
    memory_stream_json,
    entity_knowledge_store_json,
    system_persona_txt,
    user_persona_txt,
    past_chat_json,
)

For multi-agent scenarios using FalkorDB, you can create separate agents with unique user_ids:

# User A personal agent
chat_agent_user_a = ChatAgent(
    "Personal Agent",
    memory_stream_json_user_a,
    entity_knowledge_store_json_user_a,
    system_persona_txt_user_a,
    user_persona_txt_user_a,
    past_chat_json_user_a,
    user_id='user_a_id'
)

You can also easily add and remove custom tools:

def multiply(a: int, b: int) -> int:
    """Multiply two integers and returns the result integer"""
    return a * b

chat_agent.add_tool({"multiply": multiply})
chat_agent.remove_tool("multiply")

Why Use Memary

Memary is built on core principles designed to enhance agent intelligence with minimal developer effort.

• Auto-generated Memory: Agent memory automatically updates as the agent interacts, capturing all memories for display and analysis. It supports combining different databases seamlessly.
• Memory Modules: Tracks user preferences and knowledge depth through the Memory Stream (breadth of knowledge) and Entity Knowledge Store (depth of knowledge), offering insights into user interests and learning.
• System Improvement: Mimics how human memory evolves and learns over time, providing metrics on agent improvement.
• Rewind Memories: Keeps track of all chats, allowing users to rewind agent executions and access memory at specific periods (feature coming soon).
• Knowledge Graphs: Utilizes graph databases to store knowledge, employing recursive and multi-hop retrieval for efficient information access and reduced latency.
• New Context Window: Tailors agent responses by integrating agent responses, most relevant entities, and summarized chat history to match the user's current interests and knowledge level.

Links

• GitHub Repository
• Documentation
• Demo Video
• PyPI Package
• LinkedIn
• Follow on X