---

Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents

MemoryAgentJev得克萨斯大学达拉斯分校

Jiang D, Li Y, Li B. Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents. 2026.

https://github.com/libingzheren/Jev-Mem

---

Jev-Mem：面向高效 AI 智能体的系统一控制式智能体记忆

Abstract

Agentic memory is becoming essential for long-horizon AI agents, yet many existing systems rely on autoregressive LLMs to control how memories are organized, retrieved, and used, placing expensive generation on the critical path of memory operations. We introduce Jev-Mem, a new agentic memory architecture inspired by System-One/System-Two cognition. System One captures fast, lightweight decision-making, whereas System Two performs slower, deliberative reasoning. Jev-Mem brings this division of labor to agentic memory through a dedicated System-One control plane, a structured multi-relational memory plane, and a System-Two reasoning plane.

智能体记忆正成为长程 AI 智能体不可或缺的组成部分，但许多现有系统依赖自回归大语言模型来控制记忆的组织、检索与使用方式，使昂贵的生成过程处于记忆操作的关键路径上。 我们提出 Jev-Mem，这是一种受系统一／系统二认知机制启发的新型智能体记忆架构。 系统一负责快速、轻量的决策，而系统二执行较慢、需要深思熟虑的推理。 Jev-Mem 通过专用的系统一控制平面、结构化的多关系记忆平面和系统二推理平面，将这种分工引入智能体记忆。

The System-One controller governs memory typing and relational organization during construction, and dynamically performs query routing, retrieval-budget allocation, graph traversal, candidate scoring, and adaptive stopping during retrieval. System Two is invoked only for complex reasoning and answer synthesis. This design improves both memory effectiveness and system efficiency: on LoCoMo Jev-Mem achieves an overall LLM-as-a-Judge score of 0.777, an 11.0% relative improvement over the strongest baseline, while reducing memory construction time to 158 s, a 6.6× speedup over the fastest competing memory system, and lowering average query latency to 0.93 s, a 36.7% reduction. The code of Jev-Mem is publicly available.

系统一控制器在记忆构建阶段负责记忆类型判定与关系组织，并在检索阶段动态执行查询路由、检索预算分配、图遍历、候选评分和自适应停止。 只有在需要复杂推理和答案综合时，才会调用系统二。 这一设计同时提高了记忆效果与系统效率：在 LoCoMo 上，Jev-Mem 的总体 LLM-as-a-Judge 评分达到 0.777，比最强基线相对提升 11.0%；同时将记忆构建时间缩短至 158 秒，相较构建速度最快的竞争记忆系统加速 6.6 倍，并将平均查询延迟降低至 0.93 秒，降幅为 36.7%。 Jev-Mem 的代码已公开。

1. Introduction

Large language model (LLM) agents are increasingly expected to operate as persistent systems over long interaction horizons, supporting applications such as coding assistants, personal agents, research agents, and autonomous workflows. Such agents typically interleave reasoning with tool invocation and environment actions, and operate over multi-step environments such as interactive web platforms and software repositories. During these interactions, agents continuously accumulate user preferences, task history, and environment knowledge, quickly exceeding what can be maintained within a fixed context window. Enlarging the nominal context length does not by itself resolve this problem, since models do not reliably exploit all positions of a long input, and long-term interactive settings introduce additional indexing, retrieval, and reading challenges.

人们越来越期待大语言模型（LLM）智能体能够作为持久运行的系统，开展长期交互，支持编程助手、个人智能体、研究智能体和自主工作流等应用。 这类智能体通常交替进行推理、工具调用和环境动作，并在交互式 Web 平台、软件仓库等多步环境中运行。 在这些交互中，智能体不断积累用户偏好、任务历史和环境知识，很快就会超出固定上下文窗口所能容纳的范围。 仅仅扩大标称上下文长度并不能解决这一问题，因为模型无法可靠地利用长输入中所有位置的信息，而长期交互场景还会带来额外的索引、检索和阅读挑战。

To remain effective over time, agents therefore need mechanisms that can retain useful experience beyond the prompt and recover it when needed. This requirement has made agentic memory which has the ability to preserve, organize, update, and retrieve past experience.

因此，为了长期保持有效性，智能体需要相应机制，将有用经验保存在提示词之外，并在需要时重新取回。 这一需求催生了能够保存、组织、更新和检索过往经验的 智能体记忆。

Recent work has transformed agentic memory from passive storage into an active and structured component of LLM agents. Early systems mainly store past interactions and retrieve them through semantic similarity, while newer approaches selectively retain salient information, consolidate repeated observations, and organize memory across hierarchical stores. Procedural memory further captures reusable skills, distilled experience, and recurring workflows. More recent relational approaches represent memories through knowledge graphs or multiple graph views, enabling retrieval over semantic, temporal, causal, and entity relationships rather than isolated memory fragments.

近期研究已将智能体记忆从被动存储转变为 LLM 智能体中主动、结构化的组成部分。 早期系统主要存储过往交互，并通过语义相似度进行检索；较新的方法则有选择地保留重要信息、整合重复观测，并在分层存储中组织记忆。 程序性记忆进一步记录可复用的技能、提炼后的经验以及反复出现的工作流。 更新的关系型方法通过知识图谱或多个图视图来表示记忆，使检索能够沿语义、时间、因果和实体关系展开，而不是只针对孤立的记忆片段。

As memory becomes richer, however, controlling it becomes increasingly expensive. Persistent agents must repeatedly decide what to store, update, connect, retrieve, and when to stop searching. Existing systems typically rely on either fixed heuristics or general-purpose autoregressive LLMs: the former are efficient but inflexible, while the latter provide semantic flexibility at the cost of repeated token generation. As these decisions appear throughout memory construction and retrieval, memory control itself can become a major source of latency and inference overhead.

然而，随着记忆日益丰富，控制记忆的代价也越来越高。 持久运行的智能体必须反复决定存储、更新、连接和检索哪些内容，以及何时停止搜索。 现有系统通常依赖固定启发式规则或通用自回归 LLM：前者高效但缺乏灵活性，后者具备语义灵活性，却需要付出反复生成 token 的代价。 由于这些决策贯穿记忆构建与检索过程，记忆控制本身可能成为延迟与推理开销的重要来源。

This observation leads us to reconsider agentic memory through the lens of System One and System Two, a distinction drawn from dual-process accounts of human reasoning that separate fast, automatic processing from slower and more deliberative thought. Many memory operations including memory typing, relation judgment, query routing, candidate scoring, and stopping are structured, high-frequency decisions that naturally fit lightweight System-One computation, while evidence synthesis and final answer generation remain better suited to System Two. Jev makes such a separation practical by producing typed probabilistic decisions without autoregressive generation.

这一观察促使我们从 系统一与系统二 的视角重新审视智能体记忆；该区分源自人类推理的双过程理论，将快速、自动的处理与较慢、需要更多深思熟虑的思考区分开来。 许多记忆操作，包括记忆类型判定、关系判断、查询路由、候选评分和停止决策，都是结构化的高频决策，天然适合轻量的系统一计算；而证据综合与最终答案生成仍更适合系统二。 Jev 无需自回归生成即可输出带类型的概率决策，使这种分工具备了可行性。

Yet several questions remain: How should System One be integrated across the memory lifecycle? How can a weaker reasoner control memory without hurting downstream accuracy? What memory structure and retrieval process best support lightweight control?

不过，仍有若干问题有待解答：应如何将系统一融入记忆的整个生命周期？ 推理能力较弱的模型如何在不损害下游准确率的情况下控制记忆？ 什么样的记忆结构和检索过程最适合支持轻量控制？

Inspired by this observation, we propose Jev-Mem, a new System-One/System-Two architecture that treats memory control itself as a first-class systems layer. Unlike existing agentic memory systems that rely on heuristics or repeatedly invoke autoregressive LLMs throughout memory construction and retrieval, Jev-Mem introduces a dedicated System-One control plane that handles high-frequency structured decisions, a shared structured memory data plane, and a System-Two reasoning plane reserved for complex synthesis and answer generation. This separation redesigns the full memory lifecycle: on the write path, System One performs memory typing, redundancy filtering, and semantic, temporal, causal, and entity relation construction; on the read path, it performs query routing, retrieval-budget allocation, graph traversal, candidate scoring, evidence assessment, and adaptive stopping.

受这一观察启发，我们提出 Jev-Mem，这是一种新的系统一／系统二架构，将记忆控制本身作为系统中的一等层级。 现有智能体记忆系统在整个记忆构建与检索过程中依赖启发式规则，或反复调用自回归 LLM；与之不同，Jev-Mem 引入专用的系统一控制平面来处理高频结构化决策，同时设置共享的结构化记忆数据平面，以及专用于复杂综合与答案生成的系统二推理平面。 这种分离重新设计了记忆的完整生命周期：在写入路径上，系统一执行记忆类型判定、冗余过滤，以及语义、时间、因果和实体关系构建；在读取路径上，它执行查询路由、检索预算分配、图遍历、候选评分、证据评估和自适应停止。

By moving these frequent decisions out of the generative reasoning loop, Jev-Mem reduces unnecessary autoregressive inference while still preserving strong reasoning capability through System Two. Jev provides one concrete realization of the System-One controller through typed probabilistic decisions, but the key novelty of Jev-Mem is the architectural separation of lightweight memory control from deliberative reasoning across both memory construction and retrieval.

通过将这些高频决策移出生成式推理循环，Jev-Mem 减少了不必要的自回归推理，同时仍借助系统二保留强大的推理能力。 Jev 通过带类型的概率决策提供了系统一控制器的一种具体实现，但 Jev-Mem 的关键创新在于：在记忆构建与检索两个阶段，都从架构上将轻量记忆控制与深思熟虑的推理分离。

We make four main contributions.

我们的主要贡献有四点。

• We identify an opportunity to leverage System-One-style lightweight decision-making to enable more efficient agentic memory.
• We introduce Jev-Mem, a System-One/System-Two architecture that separates lightweight structured memory control from expensive generative reasoning.
• We develop a unified System-One control plane for both memory construction and retrieval, including multi-relational organization, query routing, adaptive graph traversal, candidate scoring, and stopping.
• Jev-Mem achieves the best overall accuracy while also delivering the highest efficiency among state-of-the-art baselines.

• 我们指出，可以利用系统一式的轻量决策来实现更高效的智能体记忆。
• 我们提出 Jev-Mem，这是一种将轻量的结构化记忆控制与昂贵的生成式推理相分离的系统一／系统二架构。
• 我们为记忆构建和检索开发了统一的系统一控制平面，涵盖多关系组织、查询路由、自适应图遍历、候选评分与停止决策。
• 与最先进的基线相比，Jev-Mem 在取得最佳总体准确率的同时，也实现了最高的效率。

图1：智能体记忆的工作流程。

2. Background and Motivation

2.1. Agentic Memory

Figure 1 demonstrates the workflow of agnetic memory. Let an agent maintain an evolving memory $M_t$. At interaction step $t$, a query $q_t$ retrieves relevant evidence

图1 展示了智能体记忆的工作流程。 设智能体维护一份随时间演化的记忆 $M_t$。 在交互步骤 $t$，查询 $q_t$ 检索相关证据

$$\begin{array}{}\text{(1)}& E_t=R(q_t,M_t)\end{array}$$

which is then provided to a language model for reasoning and response generation:

随后将这些证据提供给语言模型，用于推理和生成回复：

$$\begin{array}{}\text{(2)}& o_t=L(q_t,E_t)\end{array}$$

The resulting interaction may subsequently update the memory:

由此产生的交互随后可能更新记忆：

$$\begin{array}{}\text{(3)}& M_{t+1}=U(M_t,q_t,o_t)\end{array}$$

Agentic memory has evolved from simple retrieval over stored interaction histories toward increasingly active memory management. Modern systems may selectively organize observations, infer relationships among memories, consolidate information, route queries across different memory structures, and adapt retrieval based on the current query. As a result, memory is no longer only a passive store accessed by a retrieval function. It increasingly behaves as a dynamic subsystem that continuously makes decisions about how information should be organized and accessed.

智能体记忆已经从对已存储交互历史进行简单检索，发展为日益主动的记忆管理。 现代系统能够有选择地组织观测、推断记忆之间的关系、整合信息、在不同记忆结构之间路由查询，并根据当前查询调整检索。 因此，记忆不再只是通过检索函数访问的被动存储。 它越来越像一个动态子系统，持续决定如何组织和访问信息。

This shift introduces an important systems question that has received comparatively less attention: what computational mechanism should execute these memory-management decisions? Both memory update $U$ and retrieval $R$ contain frequent semantic decisions. During memory construction, the system must determine how new information should be characterized and connected to existing knowledge. During retrieval, it must determine where to search, which candidates are useful, how much additional search is warranted, and when sufficient evidence has been collected.

这一转变引出了一个此前受到较少关注的重要系统问题：应由什么计算机制来执行这些记忆管理决策？ 记忆更新 $U$ 与检索 $R$ 都包含频繁的语义决策。 在记忆构建过程中，系统必须决定如何描述新信息的特征，以及如何将其与已有知识相连接。 在检索过程中，系统必须决定去哪里搜索、哪些候选有用、还需要进行多少额外搜索，以及何时已经收集到足够的证据。

External retrieval has long served as a mechanism for augmenting parametric language-model knowledge with non-parametric stores, through dense retrievers, retrieval-augmented pre-training, and retrieval-augmented generation. Agentic memory inherits much of this machinery, but differs in that its store is written by the agent's own interaction history and must be maintained and reorganized over time rather than fixed in advance.

长期以来，外部检索通过稠密检索器、检索增强预训练和检索增强生成等方式，利用非参数化存储来补充语言模型的参数化知识。 智能体记忆继承了其中许多机制，但不同之处在于：它的存储内容来自智能体自身的交互历史，必须随时间维护和重组，而不是预先固定。

A closely related line of work studies adaptive control in retrieval-augmented generation. Rather than applying the same retrieval procedure to every query, active and adaptive methods dynamically determine whether retrieval is needed, when additional evidence should be acquired, or which retrieval strategy a given query warrants. These results motivate viewing retrieval as a controlled decision process rather than a fixed top-k operation. Jev-Mem builds on this view, and extends the same principle from evidence acquisition to the broader memory lifecycle, including memory construction, query routing, budget allocation, graph traversal, and stopping.

一个密切相关的研究方向关注检索增强生成中的自适应控制。 主动式和自适应方法不会对每个查询采用相同的检索流程，而是动态决定是否需要检索、何时应获取更多证据，或给定查询适合采用哪种检索策略。 这些结果促使我们将检索视为受控的决策过程，而不是固定的 top-k 操作。 Jev-Mem 基于这一观点，将相同原则从证据获取扩展到更广泛的记忆生命周期，包括记忆构建、查询路由、预算分配、图遍历和停止决策。

Jev-Mem focuses on this memory-control layer. Rather than treating these decisions as incidental components embedded inside prompts or fixed heuristics, we make them an explicit part of the memory architecture.

Jev-Mem 关注的正是这一记忆控制层。 我们将这些决策明确纳入记忆架构，而不再把它们视为嵌入提示词或固定启发式规则中的附带组件。

2.2. The Opportunity for System-One Memory Control

A key observation is that many memory-control operations are semantic but not generative. During memory construction, the system may need to classify a memory or infer its relation to existing information; during retrieval, it may need to route a query, score candidates, or decide when to stop. These decisions typically produce bounded outputs such as labels, probabilities, or scores rather than free-form text.

一个关键观察是，许多记忆控制操作 具有语义性，但不具有生成性。 在记忆构建过程中，系统可能需要对记忆进行分类，或推断它与现有信息之间的关系；在检索过程中，系统可能需要路由查询、为候选评分，或决定何时停止。 这些决策通常输出标签、概率或分数等有界结果，而不是自由形式的文本。

Using an autoregressive LLM for such high-frequency decisions is therefore unnecessarily expensive: even simple judgments require token-by-token generation, formatting, and parsing. Because these operations repeatedly appear on the memory critical path, their overhead can accumulate quickly.

因此，使用自回归 LLM 执行这类高频决策会产生不必要的高成本：即使简单判断也需要逐 token 生成、格式化和解析。 由于这些操作反复出现在记忆的关键路径上，其开销可能迅速累积。

More broadly, cost-aware LLM inference systems have shown that different requests need not receive identical amounts of model computation. Cascading and routing approaches dynamically allocate requests across models of differing cost to improve the cost–quality tradeoff, and speculative decoding uses a small model to draft tokens that a larger model only verifies. Jev-Mem applies a related systems principle at a finer granularity: rather than routing only complete user requests between models, it separates frequent, bounded memory-control decisions from open-ended generative and deliberative reasoning.

更广泛地说，具有成本意识的 LLM 推理系统已经表明，不同请求并不需要获得相同数量的模型计算。 级联与路由方法在成本不同的模型之间动态分配请求，以改善成本与质量的权衡；推测解码则使用小模型起草 token，大模型只负责验证。 Jev-Mem 在更细的粒度上应用了类似的系统原则：它不仅在模型之间路由完整的用户请求，还将频繁、有界的记忆控制决策与开放式生成和深思熟虑的推理分离。

This creates a natural opportunity for a System-One/System-Two design: use lightweight structured prediction for frequent memory-control decisions, while reserving System Two for complex reasoning and answer synthesis.

这自然为系统一／系统二设计创造了机会：用轻量的结构化预测执行频繁的记忆控制决策，将系统二保留给复杂推理与答案综合。

图2：Jev-Mem 概览。轻量的系统一控制平面管理记忆构建与自适应检索，系统二则专用于复杂推理与答案综合。

3. Jev-Mem Algorithm Design

Jev-Mem redesigns agentic memory around a clear separation between fast memory control and deliberative reasoning. As illustrated in Figure 2, the architecture consists of three components: a System-One controller, a shared memory data plane, and a System-Two reasoning model. Rather than relying on an autoregressive LLM to manage every memory operation, the System-One controller makes lightweight, structured decisions throughout both writing and retrieval. The memory plane maintains canonical observations together with semantic, temporal, causal, and entity relations, while System Two is reserved for answer synthesis and other operations that require open-ended generation or deeper reasoning.

Jev-Mem 通过明确区分快速记忆控制与深思熟虑的推理，重新设计了智能体记忆。 如 图2 所示，该架构由三个组件构成：系统一控制器、共享记忆数据平面，以及系统二推理模型。 系统一控制器在写入和检索的整个过程中做出轻量、结构化的决策，而不依赖自回归 LLM 来管理每一次记忆操作。 记忆平面维护规范化观测及其语义、时间、因果和实体关系；系统二则专用于答案综合，以及其他需要开放式生成或更深层推理的操作。

This architecture unifies memory construction and retrieval under the same control mechanism. On the write path, Jev-Mem preserves observations, assigns memory types, selects candidate memories, and determines how new observations should connect to the existing multi-relational memory. On the read path, it routes each query to the most relevant relational views, allocates retrieval effort, checks evidence sufficiency, expands the graph when necessary, and scores newly discovered candidates in an iterative feedback loop. By treating both workflows as coordinated System-One control processes, Jev-Mem replaces a collection of isolated heuristics and repeated LLM calls with a single, adaptive memory-management architecture.

这一架构将记忆构建与检索统一在同一套控制机制之下。 在写入路径上，Jev-Mem 保存观测、指定记忆类型、选择候选记忆，并确定新观测应如何连接到现有的多关系记忆。 在读取路径上，它将每个查询路由到最相关的关系视图，分配检索投入，检查证据是否充分，在必要时扩展图，并在迭代反馈循环中对新发现的候选进行评分。 通过将两条工作流视为相互协调的系统一控制过程，Jev-Mem 用一个统一、自适应的记忆管理架构，替代了一组孤立的启发式规则和反复的 LLM 调用。

Let an observation be

将一次观测记为

$$\begin{array}{}\text{(4)}& o_t=(x_t,{\tau }_t,{\mu }_t)\end{array}$$

where $x_t$ denotes its content, ${\tau }_t$ an optional timestamp, and ${\mu }_t$ its provenance. Jev-Mem maintains

其中，$x_t$ 表示观测内容，${\tau }_t$ 为可选的时间戳，${\mu }_t$ 为其来源信息。 Jev-Mem 维护

$$\begin{array}{}\text{(5)}& {\mathcal{M}}_t=(V_t,\{E_t^g{\}}_{g\in \mathcal{G}},I_t^{\mathrm{vec}},I_t^{\mathrm{lex}})\end{array}$$

where

其中

$$\begin{array}{}\text{(6)}& \begin{array}{rl}\mathcal{G}=\{& \mathrm{semantic},\mathrm{temporal},\\ & \mathrm{causal},\mathrm{entity}\}\end{array}\end{array}$$

All relational views share the same canonical memory nodes $V_t$; multiple typed edges may connect the same pair of memories, while the vector and lexical indexes provide complementary entry points into the same memory space.

所有关系视图共享相同的规范化记忆节点 $V_t$；多种类型的边可以连接同一对记忆，而向量索引与词法索引为同一个记忆空间提供互补的入口。

3.1. Typed System-One Memory Control

The central abstraction of Jev-Mem is a typed System-One controller $\mathcal{J}(S,\mathcal{Q})$, where $S$ is structured state and $\mathcal{Q}$ is a batch of explicit decision questions. The controller produces either probabilities for independent propositions or a distribution over mutually exclusive alternatives. For example, it can estimate whether two memories are semantically related, whether a candidate is relevant to the current query, or whether the retrieved evidence is sufficient. When exactly one action is required, it selects among predefined alternatives such as temporal relations.

Jev-Mem 的核心抽象是带类型的系统一控制器 $\mathcal{J}(S,\mathcal{Q})$，其中 $S$ 为结构化状态，$\mathcal{Q}$ 为一批明确的决策问题。 控制器输出相互独立命题的概率，或互斥选项上的概率分布。 例如，它可以估计两条记忆是否在语义上相关、某个候选是否与当前查询相关，或检索到的证据是否充分。 当必须选择且仅选择一个动作时，它会从预定义选项中进行选择，例如选择时间关系。

This interface is intentionally different from free-form LLM prompting. Each decision exposes a small, known output space, allowing Jev-Mem to represent memory control directly as probabilities rather than generating intermediate natural-language reasoning and subsequently parsing it. Moreover, decisions sharing the same state can be evaluated together in a single batched invocation.

这一接口被有意设计为不同于自由形式的 LLM 提示。 每项决策都暴露一个小规模、已知的输出空间，使 Jev-Mem 能够直接用概率表示记忆控制，而不必先生成中间的自然语言推理，再对其进行解析。 此外，共享相同状态的决策可以在一次批量调用中一并评估。

The same interface governs both the write and read paths. On the write path, it controls memory typing and relation construction. On the read path, it performs query routing, candidate evaluation, and stopping. This shared control plane is a key architectural property of Jev-Mem: memory is not constructed by one collection of heuristics and retrieved by another, but instead managed throughout its lifecycle through the same System-One decision abstraction.

同一接口同时管理写入路径与读取路径。 在写入路径上，它控制记忆类型判定与关系构建。 在读取路径上，它执行查询路由、候选评估和停止决策。 这一共享控制平面是 Jev-Mem 的关键架构特征：记忆并非由一组启发式规则构建、再由另一组规则检索，而是在整个生命周期中通过同一套系统一决策抽象进行管理。

3.2. System-One-Guided Memory Construction

For each valid observation, Jev-Mem creates one canonical memory node and selectively constructs relations around it. The current design preserves observations rather than making an irreversible learned store-or-discard decision at ingestion time. This prevents information that appears unimportant initially from being permanently lost before a future query reveals its relevance. Selectivity is instead introduced when memory structure is constructed and later when memory is retrieved.

对于每条有效观测，Jev-Mem 创建一个规范化记忆节点，并有选择地构建它周围的关系。 当前设计保留观测，而不是在信息写入时通过学习模型做出不可逆的存储或丢弃决策。 这样可以防止最初看似不重要的信息，在未来查询揭示其相关性之前就永久丢失。 选择性则体现在记忆结构的构建阶段，以及之后的记忆检索阶段。

The controller first predicts four overlapping memory characteristics:

控制器首先预测四种可以相互重叠的记忆特征：

$$\begin{array}{}\text{(7)}& \begin{array}{rl}\mathbf{t}(v)=(& t_{\mathrm{episodic}},t_{\mathrm{semantic}},\\ & t_{\mathrm{procedural}},t_{\mathrm{preference}})\end{array}\end{array}$$

These scores annotate the node rather than assigning it to a single mutually exclusive category. The canonical node retains the original observation, provenance, timestamp, embedding, entities, and type scores.

这些分数用于标注节点，而不是将节点分配到单一、互斥的类别。 规范化节点保留原始观测、来源信息、时间戳、嵌入、实体和类型分数。

A central challenge is determining how a new observation relates to an increasingly large memory. Comparing it against every existing node would make controller cost grow directly with memory size. Jev-Mem therefore separates candidate discovery from relation judgment. Deterministic retrieval first combines vector similarity, lexical overlap, shared entities, and temporal proximity to identify at most $K_w$ candidates:

一个核心挑战是，如何确定新观测与不断增长的记忆之间的关系。 如果将新观测与每个已有节点进行比较，控制器成本就会直接随记忆规模增长。 因此，Jev-Mem 将 候选发现 与 关系判断 分离。 确定性检索首先结合向量相似度、词法重叠、共享实体和时间邻近性，找出至多 $K_w$ 个候选：

$$\begin{array}{}\text{(8)}& C(v)={\mathrm{TopK}}_{u\in V_t}{s}_{\mathrm{cand}}(v,u)\end{array}$$

The System-One controller then evaluates only these candidate pairs.

系统一控制器随后仅评估这些候选对。

For every pair $(v,u)$, Jev estimates semantic relatedness, directional causal influence, same-episode membership, and, when necessary, entity equivalence. Whenever reliable structured information is already available, Jev-Mem avoids unnecessary learned inference. Timestamp ordering directly creates temporal relations, while exact shared identifiers directly create entity relations. When temporal order is implicit, the controller instead chooses among before, after, during, contains, overlaps, same_time, and unknown.

对于每一对 $(v,u)$，Jev 估计语义相关性、有向因果影响、是否属于同一事件，以及必要时的实体等价性。 只要已有可靠的结构化信息，Jev-Mem 就会避免不必要的学习模型推断。 时间戳顺序直接用于创建时间关系，完全一致的共享标识符则直接用于创建实体关系。 当时间顺序是隐含的，控制器才从 before、after、during、contains、overlaps、same_time 和 unknown 中选择。

An inferred edge of type $g$ is inserted only when

只有满足下式时，才会插入推断得到的 $g$ 类型边：

$$\begin{array}{}\text{(9)}& P(g\mid v,u)\ge {\theta }_{\mathrm{rel}}\end{array}$$

Because relations are independent views over a common memory space, the same pair of nodes may simultaneously exhibit semantic, temporal, causal, and entity relationships. This differs from assigning each memory to a single graph or duplicating the same observation across several memory stores. The write path is

由于各种关系是同一记忆空间上的独立视图，同一对节点可以同时具有语义、时间、因果和实体关系。 这不同于将每条记忆分配到单一图中，也不同于将同一观测复制到多个记忆存储中。 写入路径为

$$\begin{array}{}\text{(10)}& \begin{array}{rl}{o}_t& \to \mathrm{type}\to \mathrm{candidates}\\ & \to \mathrm{relations}\to {\mathcal{M}}_t\end{array}\end{array}$$

This design improves memory quality in two ways. First, preserving canonical observations avoids premature information loss. Second, selective relation construction suppresses unnecessary graph connectivity while retaining relations that can later support semantic, temporal, causal, or entity-based reasoning.

这一设计从两个方面提高了记忆质量。 首先，保留规范化观测可以避免过早丢失信息。 其次，选择性关系构建能够抑制不必要的图连接，同时保留之后可能支持语义、时间、因果或实体推理的关系。

Periodic maintenance further examines a bounded neighborhood for redundancy, contradiction, obsolescence, and useful additional links. These decisions enrich the memory structure without removing the original evidence. If a higher-level textual abstraction is required, the control plane may explicitly escalate an approved merge or promotion to System Two; generation is therefore an optional consequence of a structured control decision rather than the default mechanism for memory maintenance.

周期性维护还会在一个有界邻域内检查冗余、矛盾、过时信息以及有用的额外连接。 这些决策丰富了记忆结构，同时保留原始证据。 如果需要更高层次的文本抽象，控制平面可以显式地将已获准的合并或提升操作交由系统二处理；因此，生成是结构化控制决策可能触发的后续操作，而非记忆维护的默认机制。

3.3. Adaptive System-One Retrieval

Jev-Mem treats retrieval as a closed-loop control process rather than a single top-k search. Given a query $q$, the controller first predicts the relevance of each relational view,

Jev-Mem 将检索视为闭环控制过程，而不是一次 top-k 搜索。 给定查询 $q$，控制器首先预测每种关系视图的相关性，

$$\begin{array}{}\text{(11)}& \mathbf{p}(q)=\{p_g(q){\}}_{g\in \mathcal{G}}\end{array}$$

together with a multi-hop requirement $h(q)$ and a recency importance score $r(q)$. Because these probabilities are evaluated independently, a query may activate several graph views simultaneously instead of being assigned to one discrete retrieval intent.

并同时预测多跳需求 $h(q)$ 与信息新近程度的重要性分数 $r(q)$。 由于这些概率是独立评估的，一个查询可以同时激活多个图视图，而不必被归入某一种离散的检索意图。

A relation type is active when

满足下式时，一种关系类型被激活：

$$\begin{array}{}\text{(12)}& p_g(q)\ge {\theta }_{\mathrm{act}}\end{array}$$

Given a total graph-expansion budget $B$, Jev-Mem distributes search effort according to the predicted graph needs:

给定图扩展总预算 $B$，Jev-Mem 根据预测的图需求分配搜索投入：

$$\begin{array}{}\text{(13)}& w_g(q)=\frac{p_g(q{)}^{\gamma }}{\sum _{j\in \mathcal{A}(q)}{p}_j(q{)}^{\gamma }}\end{array}$$

where $\mathcal{A}(q)$ denotes the active graphs and $\gamma$ controls the concentration of the allocation. After assigning a feasible minimum budget $m$ to active graphs, the remaining budget is distributed proportionally:

其中，$\mathcal{A}(q)$ 表示被激活的图，$\gamma$ 控制预算分配的集中程度。 为各个已激活的图分配可行的最低预算 $m$ 后，将剩余预算按比例分配：

$$\begin{array}{}\text{(14)}& \begin{array}{rl}{b}_g=& m+{\mathrm{LRound}}_g\\ & [(B-m|\mathcal{A}(q)|)w_g(q)]\end{array}\end{array}$$

The multi-hop prediction further determines the allowed traversal depth,

多跳预测进一步决定允许的遍历深度，

$$\begin{array}{}\text{(15)}& \begin{array}{c}D(q)=min\{D_{max},\\ max(1,\lceil D_{max}h(q)\rceil )\}\end{array}\end{array}$$

This probabilistic routing mechanism serves two purposes. It avoids spending equal retrieval effort on relations that are unlikely to help the current query, while still allowing multiple forms of evidence to be explored when a question requires them.

这一概率路由机制有两个作用。 它避免向那些不太可能有助于当前查询的关系投入同等的检索资源，同时仍允许在问题需要时探索多种形式的证据。

Anchor retrieval. Before graph traversal, Jev-Mem identifies high-quality entry points using both semantic and lexical retrieval. Vector and keyword rankings are fused through reciprocal-rank fusion:

锚点检索。 在图遍历之前，Jev-Mem 同时使用语义检索与词法检索，寻找高质量的入口节点。 向量排名与关键词排名通过倒数排名融合进行合并：

$$\begin{array}{}\text{(16)}& \begin{array}{c}{s}_{\mathrm{RRF}}(v,q)=\\ \sum _{\begin{array}{c}L\in \{L_{\mathrm{vec}},L_{\mathrm{lex}}\}\\ v\in L\end{array}}\frac{1}{\kappa +{\mathrm{rank}}_L(v)}\end{array}\end{array}$$

where $\kappa =60$. The highest-ranked nodes initialize the visited set and search frontier. Hybrid anchors provide robust starting points before the controller begins more expensive graph reasoning.

其中 $\kappa =60$。 排名最高的节点用于初始化已访问集合与搜索前沿。 在控制器开始代价更高的图推理之前，混合锚点提供了稳健的起点。

Evidence-guided expansion. After each retrieval round $d$, Jev-Mem evaluates the current evidence set $E_d$. Instead of blindly continuing until a fixed graph depth or node count is reached, the controller estimates

证据引导的扩展。 在每一轮检索 $d$ 之后，Jev-Mem 都会评估当前证据集合 $E_d$。 控制器不会盲目地持续搜索，直到达到固定的图深度或节点数量，而是估计以下量：

$$\begin{array}{}\text{(17)}& s_d& :\text{evidence sufficiency},\text{(18)}& u_d& :\begin{array}{c}\text{expected utility of}\\ \text{further retrieval}\end{array},\text{(19)}& m_d& :\text{missing required evidence},\text{(20)}& c_d& :\text{unresolved contradiction}.\end{array}$$

Retrieval terminates with sufficient evidence when

当满足下式时，检索因证据充分而终止：

$$\begin{array}{}\text{(21)}& s_d\ge {\theta }_{\mathrm{suff}}\wedge m_d<{\theta }_{\mathrm{cont}}\wedge c_d<{\theta }_{\mathrm{cont}}\end{array}$$

It may also terminate when

检索也可能在满足下式时终止：

$$\begin{array}{}\text{(22)}& u_d<{\theta }_{\mathrm{cont}}\end{array}$$

indicating that additional traversal is unlikely to improve the evidence even if the current evidence is not fully sufficient.

这表明，即使当前证据尚不完全充分，继续遍历也不太可能改善证据。

This stopping mechanism is important for both efficiency and retrieval quality. Stopping too early risks missing necessary evidence, whereas uncontrolled expansion introduces irrelevant memories that can distract downstream reasoning. Jev-Mem therefore explicitly models both evidence completeness and the expected benefit of further search.

这一停止机制对于效率和检索质量都很重要。 过早停止可能遗漏必要证据，而不受控的扩展则会引入无关记忆，干扰下游推理。 因此，Jev-Mem 显式建模证据的完整性，以及继续搜索的预期收益。

When additional evidence is needed, Jev-Mem expands neighboring nodes under the relation-specific budgets $b_g$ and global limits on nodes, edges, depth, controller calls, and latency. Each candidate $v$ is then evaluated by the System-One controller along four complementary dimensions: query relevance $a_v$, relation usefulness ${\ell }_v$, information novelty $n_v$, and support for the current evidence $c_v$.

当需要更多证据时，Jev-Mem 在各关系专属预算 $b_g$ 以及节点数、边数、深度、控制器调用次数和延迟的全局限制下，扩展相邻节点。 随后，系统一控制器沿四个互补维度评估每个候选 $v$：查询相关性 $a_v$、关系有用性 ${\ell }_v$、信息新颖性 $n_v$，以及对当前证据的支持程度 $c_v$。

These System-One predictions are combined with deterministic retrieval signals. For a candidate reached through graph type $g$, the transition score is

这些系统一预测会与确定性的检索信号相结合。 对于通过 $g$ 类型图到达的候选，其转移分数为

$$\begin{array}{}\text{(23)}& \begin{array}{c}s(v\mid q,E_d)=\\ \frac{\begin{array}{c}{\lambda }_1{z}_v+{\lambda }_2{a}_v+{\lambda }_3{p}_g(q){\ell }_v\\ +{\lambda }_4{n}_v+{\lambda }_5({\pi }_e+c_v)/2\end{array}}{\sum _{i=1}^5{\lambda }_i}\end{array}\end{array}$$

where $z_v$ denotes embedding similarity and ${\pi }_e$ is the stored edge weight. If timestamps are available, the score is adjusted using the query-specific recency prediction:

其中，$z_v$ 表示嵌入相似度，${\pi }_e$ 为存储的边权重。 如果存在时间戳，则利用针对当前查询的新近程度预测来调整分数：

$$\begin{array}{}\text{(24)}& {\rho }_v=\frac{1}{1+max(0,{\tau }_{\ast }-{\tau }_v)/\mathrm{day}}\end{array}$$$$\begin{array}{}\text{(25)}& \tilde{s}(v)=\frac{s(v)+0.1r(q){\rho }_v}{1+0.1r(q)}\end{array}$$

The top $W$ candidates form the next beam and are added to the accumulated evidence. Retrieval therefore follows an iterative feedback loop:

得分最高的 $W$ 个候选构成下一轮束搜索的候选集合，并被加入累积证据中。 因此，检索遵循一个迭代反馈循环：

$$\begin{array}{}\text{(26)}& \begin{array}{c}\text{route}\to \text{retrieve}\to \text{assess}\\ \to \text{expand}\to \text{reassess}\end{array}\end{array}$$

This process differs fundamentally from fixed top-k retrieval or static graph traversal. The relational views being explored, the allocation of search effort, the candidates retained at each round, and the decision to continue all depend on the query and on the evidence already collected. In this sense, retrieval itself becomes a System-One-controlled decision process.

这一过程与固定的 top-k 检索或静态图遍历有根本区别。 探索哪些关系视图、如何分配搜索投入、每轮保留哪些候选，以及是否继续搜索，都取决于查询和已经收集到的证据。 从这个意义上说，检索本身成为了由系统一控制的决策过程。

Once retrieval terminates, the highest-scoring $K$ memories are passed to the System-Two LLM:

检索终止后，得分最高的 $K$ 条记忆会被传递给系统二 LLM：

$$\begin{array}{}\text{(27)}& y=\mathrm{SystemTwo}(q,E)\end{array}$$

System Two performs the final synthesis over the selected evidence but is not involved in normal graph routing, candidate expansion, or stopping decisions. Jev-Mem therefore concentrates expensive autoregressive reasoning where it is most valuable while using fast structured control to improve the quality and efficiency of the evidence supplied to it.

系统二根据选定证据进行最终综合，但不参与常规的图路由、候选扩展或停止决策。 因此，Jev-Mem 将昂贵的自回归推理集中在最有价值的地方，同时利用快速的结构化控制，提高提供给系统二的证据质量与获取效率。

The control overhead remains explicitly bounded. A normal write requires one batched typing request and, when candidates exist, one batched relation request over at most $K_w$ memory pairs. A query requires one routing request and, for each retrieval round, at most one evidence assessment and one batched candidate-scoring request. Independent limits on graph expansions, inspected edges, visited nodes, depth, controller invocations, and elapsed time prevent the control process itself from growing without bound.

控制开销始终受到明确约束。 一次常规写入需要一次批量类型判定请求；存在候选时，还需要针对至多 $K_w$ 个记忆对进行一次批量关系请求。 一次查询需要一次路由请求，并在每轮检索中至多进行一次证据评估和一次批量候选评分请求。 对图扩展次数、检查的边数、访问的节点数、深度、控制器调用次数和耗时分别设置独立限制，可以防止控制过程本身无限增长。

4. Experiments

We conduct comprehensive experiments to evaluate Jev-Mem along two dimensions: reasoning effectiveness and system efficiency. We examine whether System-One-controlled memory construction and retrieval improve long-horizon question answering, and whether separating fast memory control from System-Two reasoning reduces the overhead of building and accessing persistent memory. Accordingly, we report answer accuracy together with memory construction time and per-query latency.

我们开展了全面实验，从推理效果与系统效率两个维度评估 Jev-Mem。 我们考察由系统一控制的记忆构建与检索是否能够改善长程问答，以及将快速记忆控制与系统二推理分离，是否能够降低持久记忆构建和访问的开销。 因此，我们同时报告答案准确率、记忆构建时间和每次查询的延迟。

4.1. Experimental Setup

Datasets. We evaluate long-term conversational memory on two widely used benchmarks. LoCoMo contains ultra-long multi-session conversations and evaluates an agent's ability to recover information requiring temporal, causal, and cross-session reasoning.

数据集。 我们在两个广泛使用的基准上评估长期对话记忆。 LoCoMo 包含超长的多会话对话，用于评估智能体在需要时间、因果和跨会话推理时恢复信息的能力。

Baselines. We compare Jev-Mem with representative long-term memory approaches using the same backbone answer model whenever applicable.

基线。 我们将 Jev-Mem 与有代表性的长期记忆方法比较，并在适用时使用相同的骨干回答模型。

• Full Context: provides the complete conversation history directly to the LLM without external memory.
• A-MEM : dynamically organizes and evolves memories based on agent experiences.
• Nemori : structures conversational memory through episodic segmentation and graph-based retrieval.
• MemoryOS : organizes persistent memory using a hierarchical multi-level architecture.
• MAGMA : represents memories through multiple semantic, temporal, causal, and entity relations for structured retrieval.
• Jev-Mem (ours): uses a System-One controller to construct and adaptively retrieve from a multi-relational memory, while reserving the LLM for final reasoning and answer synthesis.

• Full Context：不使用外部记忆，直接向 LLM 提供完整对话历史。
• A-MEM：根据智能体经验动态组织和演化记忆。
• Nemori：通过情节分段与基于图的检索，构建结构化的对话记忆。
• MemoryOS：使用分层的多级架构组织持久记忆。
• MAGMA：通过多种语义、时间、因果和实体关系表示记忆，以支持结构化检索。
• Jev-Mem（本文方法）：使用系统一控制器构建多关系记忆并进行自适应检索，将 LLM 保留给最终推理与答案综合。

Metrics. We evaluate Jev-Mem from both accuracy and system-efficiency perspectives. For reasoning quality, we use the LLM-as-a-Judge score, which measures whether the generated answer is correct with respect to the reference answer. For system efficiency, we report total memory construction time and average per-query latency. Memory construction time measures the end-to-end cost of building the persistent memory from the conversation history, while query latency measures the average time required to retrieve relevant evidence and produce an answer for each query. Together, these metrics capture both the effectiveness and runtime efficiency of the proposed memory architecture.

指标。 我们从准确率与系统效率两个角度评估 Jev-Mem。 对于推理质量，我们使用 LLM-as-a-Judge 评分，衡量生成的答案相对于参考答案是否正确。 对于系统效率，我们报告记忆构建总时间与平均单次查询延迟。 记忆构建时间衡量从对话历史构建持久记忆的端到端成本；查询延迟则衡量针对每个查询检索相关证据并生成答案所需的平均时间。 这些指标共同反映了所提出记忆架构的效果与运行效率。

表1：使用 LLM-as-a-Judge 指标评估的 LoCoMo 基准性能。分数越高表示性能越好。LLM 模型基于 gpt-4o-mini。

Method	Multi-Hop	Temporal	Open-Domain	Single-Hop	Adversarial	Overall
Full Context	0.468	0.562	0.486	0.630	0.205	0.481
A-MEM	0.495	0.474	0.385	0.653	0.616	0.580
MemoryOS	0.552	0.422	0.504	0.674	0.428	0.553
Nemori	0.569	0.649	0.485	0.764	0.325	0.590
MAGMA	0.528	0.650	0.517	0.776	0.742	0.700
Jev-Mem	0.623	0.637	0.618	0.802	0.962	0.777

4.2. Overall Performance

Table 1 summarizes the performance of Jev-Mem and the baselines on LoCoMo. Jev-Mem achieves the highest overall LLM-as-a-Judge score of 0.777, compared with 0.700 for the strongest baseline, corresponding to an 11.0% relative improvement. The gains are also consistent across question types: Jev-Mem performs best in five of the six categories and matches the best result on temporal reasoning.

表1 汇总了 Jev-Mem 及各基线在 LoCoMo 上的性能。 Jev-Mem 取得最高的总体 LLM-as-a-Judge 评分 0.777，而最强基线为 0.700，相当于 11.0% 的相对提升。 不同问题类型上的提升也具有一致性：Jev-Mem 在六个类别中的五个上表现最佳，并在时间推理上追平了最佳结果。

The improvement is especially clear on queries that require stronger retrieval and evidence selection. On Multi-Hop questions, Jev-Mem reaches 0.625, compared with 0.569 for the strongest baseline. On Open-Domain questions, it achieves 0.610, substantially higher than the best baseline score of 0.517. The largest gain appears in the Adversarial category, where Jev-Mem reaches 0.962, compared with 0.742 for the strongest baseline. Jev-Mem also achieves the highest Single-Hop score of 0.797 and matches the best Temporal score of 0.650.

对于需要更强检索能力和证据选择能力的查询，提升尤其明显。 在多跳问题上，Jev-Mem 达到 0.625，而最强基线为 0.569。 在开放域问题上，它达到 0.610，明显高于最佳基线的 0.517。 最大提升出现在对抗类别中，Jev-Mem 达到 0.962，而最强基线为 0.742。 Jev-Mem 还取得了最高的单跳得分 0.797，并追平了最佳时间推理得分 0.650。

These results indicate that Jev-Mem improves performance across a broad range of memory-intensive queries rather than specializing in a single reasoning pattern. The largest gains appear when the system must combine evidence across multiple memories or distinguish relevant information from plausible but misleading distractors. Jev-Mem addresses these cases through adaptive memory control: it estimates which relational views are useful for each query, allocates retrieval effort accordingly, evaluates candidate memories during traversal, and decides when the collected evidence is sufficient. As a result, retrieval is adapted to the needs of each query instead of following a fixed search strategy, allowing the reasoning model to operate on a more relevant and focused evidence set.

这些结果表明，Jev-Mem 能够改善多种记忆密集型查询的性能，而非专门针对某一种推理模式。 当系统必须结合多条记忆中的证据，或从看似合理却具有误导性的干扰信息中区分出相关信息时，提升最为明显。 Jev-Mem 通过自适应记忆控制处理这些情况：它估计哪些关系视图对每个查询有用，据此分配检索投入，在遍历过程中评估候选记忆，并判断已收集的证据何时足够充分。 因此，检索会根据每个查询的需求进行调整，而非遵循固定的搜索策略，使推理模型能够基于更相关、更聚焦的证据集合开展推理。

表2：系统效率比较：记忆构建总时间（秒）与平均查询延迟（秒）。

Method	Build Time (s)	Latency (s)
Full Context	N/A	1.74
A-MEM	3636	2.26
MemoryOS	3276	32.68
Nemori	1044	2.59
MAGMA	1404	1.47
Jev-Mem	158	0.93

4.3. Efficiency of System-One Memory Control

Table 2 compares the system efficiency of Jev-Mem with the baselines in terms of total memory construction time and average query latency. Jev-Mem achieves substantial improvements in both stages. Its total memory build time is only 158 seconds, compared with 1,044 seconds for the fastest competing memory system, corresponding to an 84.9% reduction, or a 6.6× speedup. The gap is even larger compared with A-MEM and MemoryOS, whose memory construction requires more than 3,000 seconds.

表2 从记忆构建总时间与平均查询延迟两个方面，比较了 Jev-Mem 与各基线的系统效率。 Jev-Mem 在这两个阶段都取得了显著改进。 它的记忆构建总时间仅为 158 秒，而构建速度最快的竞争记忆系统需要 1,044 秒，相当于减少 84.9%，或加速 6.6 倍。 与 A-MEM 和 MemoryOS 相比，差距更大，这两个系统的记忆构建都需要超过 3,000 秒。

Jev-Mem also achieves the lowest query latency, requiring only 0.93 seconds per query on average. This is 36.7% lower than the fastest memory-based baseline at 1.47 seconds, and 46.6% lower than directly processing the full context at 1.74 seconds. In contrast, MemoryOS requires 32.68 seconds per query, highlighting the substantial runtime overhead that memory management can introduce when complex processing remains on the retrieval path.

Jev-Mem 还实现了最低的查询延迟，每次查询平均仅需 0.93 秒。 这比最快的基于记忆的基线所需的 1.47 秒低 36.7%，比直接处理完整上下文所需的 1.74 秒低 46.6%。 相比之下，MemoryOS 每次查询需要 32.68 秒，凸显出当复杂处理仍处于检索路径上时，记忆管理可能带来的显著运行开销。

The efficiency gains reflect the separation of memory control from autoregressive reasoning in Jev-Mem. During memory construction, lightweight typed decisions are batched over a bounded candidate set rather than repeatedly invoking a general-purpose LLM for free-form memory processing. During retrieval, System-One control performs query routing, candidate evaluation, and evidence checking, while graph exploration is constrained by explicit search budgets. Adaptive stopping further avoids unnecessary traversal once sufficient evidence has been collected. Together, these mechanisms reduce the computational overhead of both constructing and accessing long-term memory.

效率提升体现了 Jev-Mem 中记忆控制与自回归推理的分离。 在记忆构建过程中，系统在有界候选集上批量执行轻量的带类型决策，而非反复调用通用 LLM 进行自由形式的记忆处理。 在检索过程中，系统一控制负责查询路由、候选评估和证据检查，而图探索受到明确搜索预算的约束。 自适应停止进一步避免了在已收集到充分证据之后仍进行不必要的遍历。 这些机制共同降低了构建和访问长期记忆的计算开销。

Notably, the efficiency improvement does not come at the expense of reasoning quality. As shown in Table 1, Jev-Mem simultaneously achieves the highest overall accuracy on LoCoMo. These results demonstrate that separating fast memory control from System-Two reasoning can improve both the effectiveness and the efficiency of agentic memory.

值得注意的是，这种效率提升并未以牺牲推理质量为代价。 如 表1 所示，Jev-Mem 同时取得了 LoCoMo 上最高的总体准确率。 这些结果表明，将快速记忆控制与系统二推理分离，可以同时提高智能体记忆的效果与效率。

5. Conclusion

We present Jev-Mem, a System-One/System-Two architecture for efficient agentic memory. Jev-Mem uses lightweight structured prediction for high-frequency memory-control decisions and reserves System Two for complex reasoning and answer synthesis. On LoCoMo, Jev-Mem achieves the best overall accuracy and efficiency among the evaluated baselines, reaching an LLM-as-a-Judge score of $0.777$, a memory construction time of $158$ s, and an average query latency of $0.93$ s. These results show that separating memory control from generative reasoning is a promising direction for building more effective and efficient long-horizon agents.

我们提出了 Jev-Mem，这是一种用于高效智能体记忆的系统一／系统二架构。 Jev-Mem 使用轻量的结构化预测来执行高频记忆控制决策，将系统二保留给复杂推理和答案综合。 在 LoCoMo 上，Jev-Mem 在所评估的基线中取得了最佳总体准确率与效率，LLM-as-a-Judge 评分达到 $0.777$，记忆构建时间为 $158$ 秒，平均查询延迟为 $0.93$ 秒。 这些结果表明，将记忆控制与生成式推理分离，是构建更有效、更高效的长程智能体的一条有前景的方向。