跳转至

文章背景与核心概要

持久化的AI内存传统上被视为一个检索问题：将先前的交互存储为原始文本，进行嵌入（embedding），并在后续查询相关上下文。尽管这种非结构化方法对主题式召回有所帮助，但对于需要准确事实、当前状态、更新、删除、聚合、关系数据、否定查询以及显式未知数的生产级AI智能体而言，往往显得力不从心。本文指出，可靠的外部AI内存必须从基于搜索的检索转变为模式接地内存（schema-grounded memory）。

论文提出了一种迭代式、模式感知的写入路径，将内存摄入系统性地分解为对象检测、字段检测和字段值提取，并辅以验证门控、局部重试和有状态的提示词控制。这种设计将解释的负载从读取路径转移到了写入路径，使得读取操作变成了对已验证记录的受限查询，而不再是对检索到的散文进行重复推理。在提取基准测试和端到端内存基准测试中，该系统（xmemory）均大幅超越了现有的前沿基准和专用内存系统，证明了对于需要稳定事实和状态计算的内存工作负载，架构设计远比单纯的检索规模或模型强度更为重要。

From Unstructured Recall to Schema-Grounded Memory: Reliable AI Memory via Iterative, Schema-Aware Extraction

arXiv: 2604.27906 [cs.AI]
Authors: Alex Petrov, Alexander Gusak, Denis Mukha, Dima Korolev
Submitted: 30 April 2026 (Last revised 13 Aug 2026, version v3)
Primary Subject: Artificial Intelligence (cs.AI)

📌 Summary

Persistent AI memory is traditionally treated as a retrieval problem: previous interactions are stored as raw text, embedded, and queried later for relevant context. While helpful for thematic recall, this unstructured approach falls short for production-grade AI agents requiring exact facts, current states, updates, deletions, aggregations, relational data, negative queries, and explicit unknowns.

This paper argues that reliable external AI memory must transition from search-based retrieval to schema-grounded memory.

Key Contributions:

  • Schema-Grounded Architecture: Schemas explicitly define what must be remembered, what can be ignored, and which values must never be inferred.
  • Iterative Write Path: Memory ingestion is systematically decomposed into object detection, field detection, and field-value extraction, supported by validation gates, local retries, and stateful prompt control.
  • Shifted Interpretation Load: By shifting interpretation from the read path to the write path, read operations become constrained queries over verified records rather than repeated inference over retrieved prose.
  • Empirical Performance:
  • Extraction Benchmark: Achieves 90.42% object-level accuracy and 62.67% output accuracy in a judge-in-the-loop configuration, outperforming frontier structured-output baselines.
  • End-to-End Memory Benchmark: The xmemory system reaches 97.10% F1 (compared to 80.16%–87.24% for third-party baselines).
  • Application-Level Tasks: xmemory reaches 95.2% accuracy, outperforming specialized memory systems, code-generated Markdown harnesses, and frontier-model application harnesses.

📚 Abstract

Persistent AI memory is often reduced to a retrieval problem: store prior interactions as text, embed them, and ask the model to recover relevant context later. This design is useful for thematic recall, but it is mismatched to the kinds of memory that agents need in production: exact facts, current state, updates and deletions, aggregation, relations, negative queries, and explicit unknowns. These operations require memory to behave less like search and more like a system of record.

This paper argues that reliable external AI memory must be schema-grounded. Schemas define what must be remembered, what may be ignored, and which values must never be inferred. We present an iterative, schema-aware write path that decomposes memory ingestion into object detection, field detection, and field-value extraction, with validation gates, local retries, and stateful prompt control. The result shifts interpretation from the read path to the write path: reads become constrained queries over verified records rather than repeated inference over retrieved prose.

We evaluate this design on structured extraction and end-to-end memory benchmarks. On the extraction benchmark, the judge-in-the-loop configuration reaches 90.42% object-level accuracy and 62.67% output accuracy, above all tested frontier structured-output baselines. On our end-to-end memory benchmark, xmemory reaches 97.10% F1, compared with 80.16%-87.24% across the third-party baselines. On the application-level task, xmemory reaches 95.2% accuracy, outperforming specialised memory systems, code-generated Markdown harnesses, and customer-facing frontier-model application harnesses. The results show that, for memory workloads requiring stable facts and stateful computation, architecture matters more than retrieval scale or model strength alone.

📋 Additional Metadata

Category Details
Comments 33 pages, 7 figures
Subjects Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
MSC Classes 68T50, 68T30, 68P20, 68P15, 94A15
ACM Classes I.2.7; I.2.4; H.3.3; H.2.1; H.2.3
DOI 10.48550/arXiv.2604.27906

🔗 Full-Text & Resources