SodaMem: Evidence-Grounded Temporal Graph Memory for LLM Agents

2026-08-08

AI Summary

This work addresses the challenges of inaccurate memory, poor temporal awareness, lack of source traceability, and limited temporal reasoning in large language model agents during extended conversations. To overcome these limitations, the authors propose an evidence-based temporal graph memory system that integrates FactEvents—structured facts annotated with mandatory source spans—into a graph architecture. By explicitly recording mention time, occurrence time, and validity intervals for each event, and by incorporating temporal relational edges such as SUPERSEDES, CONTRADICTS, and UPDATES, the system enables highly timely and auditable memory management. The approach combines typed fact extraction, hybrid lexical-dense indexing, and a planner-reader loop mechanism, leveraging DeepSeek-V4-Flash for evidence retrieval and generation. Evaluated on LongMemEval-S, it achieves 92.8% accuracy (464/500) at a cost of only $0.00161 per query, substantially outperforming existing systems.

Abstract

Large language model (LLM) agents that assist users over weeks of conversation must remember what is currently true, not merely what was once said. Flat RAG diaries and Markdown logs optimize needle retrieval but under-serve currency, provenance, and ordered temporal reasoning (Maharana et al. 2024; Wu et al. 2024; Packer et al. 2023; Chhikara et al. 2025). We present SodaMem, an evidence-grounded temporal graph memory that (i) extracts typed FactEvents with mandatory provenance spans, (ii) persists mention time, occurrence time, and validity with SUPERSEDES/CONTRADICTS/UPDATES edges under hybrid lexical-dense indexing, and (iii) answers via a planner-reader loop that gathers citable evidence before composing a final response. On LongMemEval-S, our store-of-record configuration reaches 92.8% accuracy (464/500; best of N=3) at mean $0.00161/question (approximately 18.3k tokens; median $0.00111 / approximately 14.6k) with deepseek-v4-flash. We compile public systems with estimable API cost into a cost table and cost-accuracy map; under these estimates SodaMem sits near the accuracy frontier at Flash-tier spend and strictly dominates several higher-cost, lower-accuracy points. Accuracy uses the same Flash model as reader and judge (self-grading); costs exclude ingest/judge and cross-system comparisons are compiled estimates rather than a single-harness bake-off.Our code is available at https://github.com/SodaMem/SodaMem

Problem

• temporal reasoning
• information currency
• provenance
• long-term memory
• LLM agents

Innovation

• Temporal Graph Memory
• Evidence-Grounded Reasoning
• FactEvent Extraction
• Provenance Tracking
• Planner-Reader Loop