Memcode speeds up and improves long term memory for language agents
MemCode: Discrete Semantic Representations for Long-Term Agent Memory
Artificial Intelligence
Summary
Keeping track of long conversations is hard for AI agents, which usually store bits of chat and search similar pieces later. The authors designed MemCode, which breaks down chats into small, clear pieces and stores them as discrete codes, letting the agent find related ideas even if the words differ. This method also separates quick writing from slower organizing to be more efficient. Tests showed MemCode recalls information more accurately and much faster than other systems.
What this means in practice
- •For chatbot developers: Use MemCode to enable chatbots to remember and retrieve relevant past conversations more efficiently and accurately over longer periods.
- •For voice assistant engineers: Implement MemCode to speed up and improve long-term memory recall in voice assistants managing user sessions across time.
Authors
Xin Li, Liang Hu, Duoqian Miao, Ming Peng, Qi Zhang
Abstract
Long-term memory remains a fundamental bottleneck for large language model (LLM) agents across sessions. Existing systems typically adopt a write-then-retrieve paradigm, storing context-dependent dialogue fragments and retrieving them via nearest-neighbor search in a continuous embedding space. This design yields unstable access units and systematically misses logically related but lexically distant evidence. We propose MemCode, a long-term agent memory framework that couples retrieval-oriented writing with discrete associative retrieval. MemCode rewrites dialogue into self-contained atomic memory units (AMUs), maps AMUs into a learnable codebook with isomorphic semantic quantization (ISQ), and uses codebook co-occurrence topology as a retrieval path complementary to vector search. For long-term memory stores, MemCode separates online atomic writes from asynchronous offline consolidation. On LongMemEval-S and LoCoMo with GPT-4o-mini and Qwen3-30B-A3B-Instruct-2507 backbones, MemCode achieves the best overall accuracy while keeping per-dialogue memory construction within roughly 40~130 seconds, making it 3~21 times faster than LightMem and over an order of magnitude faster than A-MEM and MemoryOS.