G-Memory introduces a hierarchical memory architecture for multi-agent systems (MAS), enabling continuous evolution through interaction. It captures generalizable insights and agent-specific collaboration trajectories using a structured graph-based design, allowing teams to adapt and improve over time by retrieving and updating relevant knowledge for new tasks. This benefits researchers and developers building adaptive MAS.

How It Works

G-Memory's core approach is a hierarchical memory structure inspired by organizational memory theory. It utilizes a graph-based design to represent and evolve memories, distinguishing between generalizable insights and agent-specific collaboration paths. When faced with a new task, the system retrieves pertinent past experiences and distilled knowledge to inform agent actions and coordination. Post-task, new interactions update the memory hierarchy, fostering adaptive learning within the MAS.

Quick Start & Requirements

• Installation: Requires Python 3.12. Setup involves creating a conda environment (conda create -n GMemory python=3.12, conda activate GMemory) and installing dependencies (pip install -r requirements.txt).
• Data: Users must download and place the ALFWorld, PDDL, and FEVER datasets in the data directory, organized into alfworld, pddl, and fever subfolders respectively.
• API Keys: OpenAI API key and base URL must be configured in a .env file, created by renaming template.env.
• Execution: Run via the ./run_mas.sh script or directly using Python commands, e.g., python tasks/run.py --task alfworld --reasoning io --mas_memory g-memory --mas_type autogen --model <your model here>.
• Links: No external links for quick-start guides or demos are provided.

Highlighted Details

• Integrates with multiple Multi-Agent System (MAS) frameworks including AutoGen, DyLAN, and MacNet.
• Supports various memory types, such as ChatDev, MetaGPT, Voyager, Generative, MemoryBank, and G-Memory itself.
• Employs a graph-based memory structure for dynamic knowledge representation and evolution.

Maintenance & Community

The provided README does not contain information regarding specific contributors, sponsorships, community channels (like Discord or Slack), or a public roadmap.

Licensing & Compatibility

The repository's license is not specified in the README, making it unclear for commercial use or integration with closed-source projects.

Limitations & Caveats

No explicit limitations, known bugs, or alpha/beta status are mentioned in the README.