A Taxonomy for the Agent Memory Zoo

What it does This repository is the living bibliography for the survey paper Memory in the Age of AI Agents. It catalogs the growing literature on how foundation-model agents store, manage, and retrieve memory, sorting papers into a unified taxonomy of Forms, Functions, and Dynamics. The authors also draw boundaries between Agent Memory and adjacent concepts like RAG and Context Engineering, which the README notes are often conflated.

The interesting bit The field is, by the authors’ own admission, a terminological free-for-all. Their response is to treat memory as a first-class primitive with its own storage media, functional roles, and operational lifecycle—an unusually architectural lens for what often gets reduced to prompt stuffing.

Key highlights

• Backs the arXiv survey (2512.13564) and was featured as Hugging Face Daily Paper #1.
• Taxonomy covers Forms (token-level, parametric, latent), Functions (factual, experiential, working), and Dynamics (formation, evolution, retrieval).
• Includes a conceptual comparison diagram distinguishing Agent Memory from LLM Memory, RAG, and Context Engineering.
• Community-maintained and open to contributions via pull requests.
• MIT licensed.

Caveats

• This is a curated reading list, not a framework or codebase—expect bibliographies, not build instructions.
• The provided sources show only the top of a very long paper list, so the full depth is implied but not fully visible here.

Verdict Worth bookmarking if you’re building agents and need to navigate the memory literature without drowning in overlapping jargon. Skip it if you are hunting for ready-to-run memory implementations.