Memory as Code: Declarative Retrieval over Codified Long-Term Memory

Key points

• Motivated by the modularity and composability of code, we propose Memory as Code (MaC), a unified memory framework that jointly designs memory construction and retrieval through a shared code abstraction.1
• MaC couples codified construction, which establishes executable memory abstractions, with declarative retrieval, which translates queries into structured retrieval programs.1
• Declarative retrieval recasts retrieval from relevance-based fact selection into declarative query processing over memory.1
• Experiments on LoCoMo and LongMemEval show relative gains of up to 15.91% in overall LLM-judge accuracy over the strongest baseline in each setting, while substantially reducing token consumption, highlighting the effectiveness of memory as code for long-horizon memory tasks.1

Sources (1)

• [1]Memory as Code: Declarative Retrieval over Codified Long-Term Memory
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 06:17 AM

  In long-term memory systems, memories are constructed from interaction histories and retrieved in response to queries, enhancing the reasoning ability of LLM agents.

  Motivated by the modularity and composability of code, we propose Memory as Code (MaC), a unified memory framework that jointly designs memory construction and retrieval through a shared code abstraction.

Extractive summary: sentences quoted from the sources.