Appendix · Memory

Back to core

Appendix: Index · Prev: World models · Next: Reasoning & planning

Overview#copy

Memory in agentic systems covers retrieval of external knowledge (documents, code, tools) and persistence of task state over long horizons. The practical question is: what is stored, when, and how it is validated.

1) Retrieval-augmented generation (RAG)#copy

Key papers#copy

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020)
• REALM (Guu et al., 2020)

Code repositories / projects#copy

• facebookresearch/DPR
• langchain-ai/langchain

2) Memory-augmented language models#copy

Key papers#copy

• RETRO: Improving language models by retrieving from trillions of tokens (Borgeaud et al., 2022)
• kNN-LM (Khandelwal et al., 2019)

Code repositories / projects#copy

• lucidrains/RETRO-pytorch
• facebookresearch/faiss

3) Agent memory as a system interface#copy

Key papers#copy

• MemGPT: Towards LLMs as Operating Systems (Packer et al., 2023)

Code repositories / projects#copy

• letta-ai/letta