Published July 17, 2026
| Version v3
Preprint
Open
Governed Enterprise AI Memory Beyond RAG: From Vector Retrieval to Permissioned Knowledge Graphs
Authors/Creators
Description
A preprint on governed shared AI memory beyond vector RAG. It separates evidence, claims, active facts, permissions, conflicts, and supersession, using Dense-Mem as an example and evaluation test bench. Version 3 documents relationship-heavy and evidence-ID retrieval attempts on a 1,000-document three-axis seed, then reports a 1,000-document six-axis smoke comparison in which evidence-first ranking with relationships reserved for discovery improved Recall@10, MRR, nDCG@10, and rank-1 rate over a retained chunk-first control.
Files
article.pdf
Additional details
Software
- Repository URL
- https://github.com/markhuangai/dense-mem
- Programming language
- Go , TypeScript
- Development Status
- Active
References
- Vaswani et al. (2017). Attention Is All You Need. arXiv:1706.03762.
- Lewis et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv:2005.11401.
- Liu et al. (2024). Lost in the Middle: How Language Models Use Long Contexts. arXiv:2307.03172.
- Edge et al. (2024). From Local to Global: A Graph RAG Approach to Query-Focused Summarization. arXiv:2404.16130.
- Hsieh et al. (2024). RULER: What's the Real Context Size of Your Long-Context Language Models? arXiv:2404.06654.
- Han et al. (2025). Retrieval-Augmented Generation with Graphs (GraphRAG). arXiv:2501.00309.
- Zhu et al. (2025). Knowledge Graph-Guided Retrieval Augmented Generation. NAACL 2025. doi:10.18653/v1/2025.naacl-long.449.
- Hong et al. (2025). Context Rot: How Increasing Input Tokens Impacts LLM Performance. Chroma Technical Report.
- McKinsey & Company (2025). The State of AI in 2025: Agents, Innovation, and Transformation.
- Stanford Institute for Human-Centered Artificial Intelligence (2026). The 2026 AI Index Report.
- Rasmussen et al. (2025). Zep: A Temporal Knowledge Graph Architecture for Agent Memory. arXiv:2501.13956.
- Taheri (2026). Governed Memory: A Production Architecture for Multi-Agent Workflows. arXiv:2603.17787.
- Chamarty (2026). Context Objects: A Temporal, Provenance-Aware Memory Primitive for Enterprise AI Agents. SSRN. doi:10.2139/ssrn.6775102.
- Srinivasan (2026). Stateless Decision Memory for Enterprise AI Agents. arXiv:2604.20158.
- Yu et al. (2025). EKRAG: Benchmark RAG for Enterprise Knowledge Question Answering. ACL Anthology.
- Cohen et al. (2025). WixQA: A Multi-Dataset Benchmark for Enterprise Retrieval-Augmented Generation. arXiv:2505.08643.
- Sun et al. (2026). EnterpriseRAG-Bench: A RAG Benchmark for Company Internal Knowledge. arXiv:2605.05253.
- Huang (2026). AI Memory Beyond RAG. Blog post. https://markhuang.ai/blog/ai-memory-beyond-rag.
- Huang (2026). Dense-Mem. Software repository. https://github.com/markhuangai/dense-mem.
- Thakur et al. (2021). BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models. NeurIPS Datasets and Benchmarks.
- Wadden et al. (2020). Fact or Fiction: Verifying Scientific Claims. EMNLP 2020. doi:10.18653/v1/2020.emnlp-main.609.
- Yang et al. (2018). HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering. EMNLP 2018. doi:10.18653/v1/D18-1259.
- Bajaj et al. (2016). MS MARCO: A Human Generated Machine Reading Comprehension Dataset. arXiv:1611.09268.
- Trivedi et al. (2022). MuSiQue: Multihop Questions via Single-hop Question Composition. Transactions of the Association for Computational Linguistics 10. doi:10.1162/tacl_a_00475.
- Dasigi et al. (2021). A Dataset of Information-Seeking Questions and Answers Anchored in Research Papers. NAACL 2021. doi:10.18653/v1/2021.naacl-main.365.
- Wu et al. (2024). LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory. arXiv:2410.10813.