Search papers, labs, and topics across Lattice.
The paper introduces LycheeMemory V2, a novel long-term memory framework for LLM agents that utilizes semantic segment-level consolidation instead of traditional turn-level consolidation. This approach significantly reduces the cost of memory construction by batching multiple interactions into segments, thus preserving contextual evidence while minimizing the frequency of LLM encoding. Experimental results demonstrate that LycheeMemory V2 achieves state-of-the-art performance on benchmark tasks while drastically lowering the number of construction tokens required, highlighting a critical trade-off in memory efficiency and accuracy.
Segment-level consolidation in LycheeMemory V2 slashes construction costs by up to 86% while maintaining high accuracy in long-term memory tasks.
Long-horizon LLM agents must preserve information from past interactions to support future tasks. Existing memory systems typically rely on eager consolidation, invoking LLMs after each interaction to extract, summarize, or update memories. This design makes memory construction increasingly costly as conversations grow. Coarse summarization can reduce construction cost but risks discarding fine-grained contextual evidence, whereas larger retrieval contexts or multi-hop LLM reasoning shift the overhead to query time. We present LycheeMemory V2, an efficient long-term memory framework that replaces turn-level consolidation with semantic segment-level consolidation. Instead of consolidating every interaction, LycheeMemory batches multiple exchanges into segments and encodes each finalized segment into context-independent typed memory records. Segment-level batching lowers LLM encoding frequency, while semantic boundary detection helps preserve coherent event-level and temporal evidence compared with fixed-window batching. The resulting records are organized with lightweight structured indexes for query-planned evidence retrieval. Experiments using GPT-4.1-Mini show that LycheeMemory achieves state-of-the-art performance, reaching 89.22% on LoCoMo and 92.20% on LongMemEval-S. Compared with A-Mem, it reduces construction tokens by 86.0% on LoCoMo and 75.9% on LongMemEval-S without increasing query-time token usage. More broadly, our results suggest that the accuracy--cost trade-off of long-term agent memory depends not only on what information is retained, but also on the granularity at which it is consolidated.
