Summary

This repository serves as a comprehensive paper list and a structured taxonomy for the rapidly evolving field of "Memory in the Age of AI Agents." It addresses the fragmentation and inconsistent terminology prevalent in current research by offering a unified framework. Targeted at AI researchers, engineers, and practitioners, it provides a clear, organized overview of memory's role in enabling long-horizon reasoning, continuous adaptation, and effective interaction for foundation model-based agents, thereby accelerating understanding and development in this critical area.

How It Works

The core contribution is a novel, unified taxonomy that categorizes agent memory through three distinct lenses: Forms (how memory is stored: Token-level, Parametric, or Latent), Functions (why agents need memory: Factual, Experiential, or Working Memory), and Dynamics (how memory operates: Formation, Evolution, and Retrieval). This structured approach aims to bridge conceptual gaps, distinguish agent memory from related concepts like RAG and Context Engineering, and establish memory as a fundamental primitive for future agentic intelligence. The repository meticulously lists and categorizes relevant research papers according to this taxonomy.

Highlighted Details

• Presents a unified taxonomy for agent memory, covering its Forms, Functions, and Dynamics, offering a novel organizational structure for the field.
• Categorizes a substantial list of research papers, providing direct links to their publications (e.g., arXiv, Hugging Face).
• Explicitly differentiates "Agent Memory" from related concepts such as Retrieval-Augmented Generation (RAG) and Context Engineering.
• Includes papers spanning various memory types, from explicit token-level storage to implicit parametric knowledge and latent states, and across different functional needs like factual recall and experiential learning.

Maintenance & Community

The project actively welcomes community contributions through pull requests, fostering collaborative development and expansion of the paper list. A link to a GitHub star chart is provided, offering a metric for community interest and project popularity.

Licensing & Compatibility

The repository is released under the MIT License. This permissive license generally allows for broad use, including integration into commercial and closed-source projects, provided attribution is maintained.

Limitations & Caveats

As a curated survey and paper list, the repository itself does not have functional limitations. However, it implicitly highlights the current state of research, which is characterized by fragmentation and a lack of standardized terminology, underscoring the need for the organizational framework it provides. The list's comprehensiveness is dependent on ongoing research and community contributions.