TsinghuaC3I/Awesome-Memory-for-Agents

What it is. This repository is a well-organized list of academic papers about "memory" in AI agents — the mechanisms that let a language model keep information across conversations or tasks. It is meant as a research reference, not a software library.

How it organizes the field. The README introduces a simple framework to sort that work:

• Short‑term memory – information that lives only inside the current request (the model's context window). It is transient and tied to one task.
• Long‑term memory – information stored outside that window so it persists across tasks. It is further split:
  • Experience – knowledge that has been explicitly validated by task outcomes (success/failure).
  • Memory – any other persistent information that is not judged by task results.

That taxonomy maps onto three application areas, shown in the table below:

Application	Memory content	What the research focuses on
Personalization	user profiles, interaction history, facts, etc.	Keeping a conversation continuous and personal, usually by retrieving relevant information from an external memory pool.
Learning from Experience	trajectories, success/failure lessons, reusable skills, etc.	Accumulating knowledge across tasks and transferring it to new situations.
Long‑horizon Agentic Task	intermediate results, reasoning traces, environmental observations, etc.	Managing context within a single long task using summarization, reflection, or a scratchpad.

Content of the paper list. The repository then lists papers under these categories. The README excerpt shows two of them:

• Personalization – many papers with titles like “MemRec”, “Memanto”, “StructMem”, etc., each linked to its arXiv or ACL page. (Dates range from 2023 to 2026.)
• Learning from Experience – similar list with titles like “ForeDreamer”, “ThinkRetrieve”, “MemGovern”, etc. (Only partially shown in the excerpt.)

The table of contents also mentions sections for Survey, Benchmark, and Product & Project, but the provided excerpt does not include the actual listing for those sections, so we cannot describe their content here.

Who made it. The list is maintained by Hongyi Liu, Yu Fu, Kaiyan Zhang, with contributions from several other named researchers (Yuxin Zuo, Che Jiang, Guoli Jia, Yuru Wang, Kaikai Zhao, Yuchen Fan, Zhenzhao Yuan, Kai Tian, Weizhi Wang).

In a nutshell. This is a curated, clearly categorized bibliography for researchers working on memory systems for AI agents — a go‑to reference for who published what and where. It does not contain code or implementations, only paper links and a thoughtful taxonomy to guide readers through the field.