Memary

What it is

Memary is described as the Open Source Memory Layer For Autonomous Agents. It provides a memory system for agents, including a Memory Stream and an Entity Knowledge Store that work with a Knowledge Graph to track and organize information from agent interactions. The project aims to enable auto-generated memory, memory module integration, and a system for analyzing agent improvement over time.

How it works

The repository combines:

• A Knowledge Graph backed by a graph database, with nodes added via Llama Index and external queries via Perplexity for certain tasks.
• A Memory Module consisting of a Memory Stream (captures entities and timestamps) and an Entity Knowledge Store (tracks frequency and recency of entity references).
• An agent layer (ReAct-based in examples) that uses a routing agent to plan and execute queries with tools such as search, vision, and location capabilities.
• A multi-graph capability via FalkorDB (multi-agent support) referenced in the latest release.

Core code locations mentioned include the ChatAgent class in src/agent, and modules for memory, knowledge graph, and benchmarks.

Getting started

Quickstart / Install

• Install memary via pip:

pip install memary

• Alternatively, install dependencies from requirements:

pip install -r requirements.txt

Run memary

The README provides steps to run the app via Streamlit after setting up environment credentials and optional FalkorDB multi-graph usage. Example workflows include creating multi-agent setups and adding custom tools to the agent. The containerized path shown for running the app is:

cd streamlit_app
streamlit run app.py

Basic usage example

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,
)

Multi-Graph usage

When using FalkorDB, memary supports multi-agent setups with user-specific agents:

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'
)

chat_agent_user_b = ChatAgent(
    "Personal Agent",
    memory_stream_json_user_b,
    entity_knowledge_store_json_user_b,
    system_persona_txt_user_b,
    user_persona_txt_user_b,
    past_chat_json_user_b,
    user_id='user_b_id'
)

Adding/Removing 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")

Recent releases

Latest release: v0.1.5 (2024-10-22) introduced FalkorDB Multi Graph ability by @galshubeli. Full changelog is available at the GitHub compare page for v0.1.4...v0.1.5. Previous releases include v0.1.4 (2024-09-09), v0.1.3 (2024-05-27), v0.1.2 (2024-05-21), and v0.1.1 (2024-05-09) with notes about docs, default models, and installation changes.

Changelog references:

• v0.1.5: FalkorDB Multi Graph ability by @galshubeli in pull/61
• v0.1.4: Diagram additions, updated descriptions, fix REA
• v0.1.3: version 0.1.2 quickfix, default model suggestions
• v0.1.2: docs: update README for pip install, Ollama integration
• v0.1.1: big update: pip install memary==0.1.1

Traction

Repo has 2636 stars and 203 forks as of now. Open issues: 14.

Behind the repo

Memary is hosted at kingjulio8238/Memary with MIT license. The README shows a quickstart and mentions Ollama-based local models and multiple external APIs (OpenAI, FalkorDB, Neo4j, Perplexity, Google Maps, Alpha Vantage).

Caveats

License: MIT. Created 2024-04-10. Last push 2024-10-22. Language: Jupyter Notebook. Topics include agents, knowledge-graph, memory, multiagent-systems, rag, self-improvement.