A field guide to curing LLM amnesia

What it does

Awesome-AI-Memory is a curated knowledge base that collects and classifies research on memory systems for large language models and agents. It organizes 402 papers and 104 open-source projects into a taxonomy covering storage mechanisms, temporal scope, memory operations, and evaluation benchmarks. The repository acts as a centralized map for a field that sits at the intersection of NLP, information retrieval, and cognitive science.

The interesting bit

Rather than treating RAG or vector databases as isolated tricks, the project frames them as part of a unified “memory system” stack with distinct storage, processing, retrieval, and control layers. It also explicitly excludes generic pre-training research, keeping the scope tightly focused on external, controllable memory that persists across sessions.

Key highlights

• Surveys 402 papers and 104 open-source projects, sorted by methods, systems, benchmarks, and surveys
• Defines a four-layer memory system architecture: storage, processing, retrieval, and control
• Covers memory types from short-term context to episodic, semantic, and procedural long-term memory
• Includes multi-agent shared memory, cognitive-science-inspired designs, and memory management policies like forgetting and compression
• Maintains explicit scope boundaries: no general pre-training or traditional databases unless they directly transfer to LLM memory

Caveats

• This is a curated bibliography and taxonomy, not a deployable memory framework or library
• The bulk of the content is academic paper indexing rather than practical implementation guides

Verdict

Worth bookmarking if you are building agent architectures or researching long-term memory for LLMs. Skip it if you are looking for a drop-in memory backend to install today.