2502.12110v11

# A-Mem: Agentic Memory for LLM Agents Abstract While large language model (LLM) agents can effectively use external tools for complex real-world tasks, they require memory systems to leverage historical experiences. Current memory systems enable basic storage and retrieval but lack sophisticated memory organization, despite recent attempts to incorporate graph databases. Moreover, these systems’ fixed operations and structures limit their adaptability across diverse tasks. To address this limitation, this paper proposes a novel agentic memory system for LLM agents that can dynamically organize memories in an agentic way. Following the basic principles of the Zettelkasten method, we designed our memory system to create interconnected knowledge networks through dynamic indexing and linking. When a new memory is added, we generate a comprehensive note containing multiple structured attributes, including contextual descriptions, keywords, and tags. The system then analyzes historical memories to identify relevant connections, establishing links where meaningful similarities exist. Additionally, this process enables memory evolution – as new memories are integrated, they can trigger updates to the contextual representations and attributes of existing historical memories, allowing the memory network to continuously refine its understanding. Our approach combines the structured organization principles of Zettelkasten with the flexibility of agent-driven decision making, allowing for more adaptive and context-aware memory management. Empirical experiments on six foundation models show superior improvement against existing SOTA baselines. <details> <summary>x1.png Details</summary>  ### Visual Description Icon/Small Image (25x24) </details> Code for Benchmark Evaluation: https://github.com/WujiangXu/AgenticMemory <details> <summary>x2.png Details</summary>  ### Visual Description Icon/Small Image (25x24) </details> Code for Production-ready Agentic Memory: https://github.com/WujiangXu/A-mem-sys 1 Introduction Large Language Model (LLM) agents have demonstrated remarkable capabilities in various tasks, with recent advances enabling them to interact with environments, execute tasks, and make decisions autonomously [23, 33, 7]. They integrate LLMs with external tools and delicate workflows to improve reasoning and planning abilities. Though LLM agent has strong reasoning performance, it still needs a memory system to provide long-term interaction ability with the external environment [35]. Existing memory systems [25, 39, 28, 21] for LLM agents provide basic memory storage functionality. These systems require agent developers to predefine memory storage structures, specify storage points within the workflow, and establish retrieval timing. Meanwhile, to improve structured memory organization, Mem0 [8], following the principles of RAG [9, 18, 30], incorporates graph databases for storage and retrieval processes. While graph databases provide structured organization for memory systems, their reliance on predefined schemas and relationships fundamentally limits their adaptability. This limitation manifests clearly in practical scenarios - when an agent learns a novel mathematical solution, current systems can only categorize and link this information within their preset framework, unable to forge innovative connections or develop new organizational patterns as knowledge evolves. Such rigid structures, coupled with fixed agent workflows, severely restrict these systems’ ability to generalize across new environments and maintain effectiveness in long-term interactions. The challenge becomes increasingly critical as LLM agents tackle more complex, open-ended tasks, where flexible knowledge organization and continuous adaptation are essential. Therefore, how to design a flexible and universal memory system that supports LLM agents’ long-term interactions remains a crucial challenge. <details> <summary>x3.png Details</summary>  ### Visual Description ## Diagram: LLM Agent Interaction ### Overview The image is a diagram illustrating the interaction between an environment, LLM (Large Language Model) agents, and memory. It shows a cyclical flow of information between these three components. ### Components/Axes * **Environment:** Represented by a globe icon with blue oceans and green landmasses. * **LLM Agents:** Represented by a blue robot icon inside a speech bubble. * **Memory:** Represented by a stack of three purple rectangles. * **Interaction:** A bidirectional arrow connecting the Environment and LLM Agents. * **Write:** A unidirectional arrow pointing from LLM Agents to Memory. * **Read:** A unidirectional arrow pointing from Memory to LLM Agents. ### Detailed Analysis * **Environment:** The globe icon suggests a real-world or simulated environment that the LLM agents interact with. * **LLM Agents:** The robot icon represents the LLM agents, which are the central processing units in this diagram. They interact with the environment and the memory. * **Memory:** The stack of rectangles represents a memory storage system where the LLM agents can write and read data. * **Interaction:** The bidirectional arrow labeled "Interaction" indicates that the LLM agents can both receive information from and send information to the environment. * **Write:** The unidirectional arrow labeled "Write" indicates that the LLM agents can store data in the memory. * **Read:** The unidirectional arrow labeled "Read" indicates that the LLM agents can retrieve data from the memory. ### Key Observations * The diagram shows a closed-loop system where the LLM agents interact with the environment, write data to memory, read data from memory, and then use that data to further interact with the environment. * The LLM Agents are central to the diagram, acting as the intermediary between the Environment and Memory. ### Interpretation The diagram illustrates a basic architecture for an LLM agent system. The LLM agents interact with an environment, learn from it, and store information in memory. This stored information can then be used to improve future interactions with the environment. The "Write" and "Read" operations highlight the LLM's ability to learn and adapt over time. The diagram suggests a continuous learning process where the LLM agents are constantly updating their knowledge based on their interactions with the environment and the data stored in memory. This model is applicable to various applications, such as robotics, natural language processing, and decision-making systems. </details> (a) Traditional memory system. <details> <summary>x4.png Details</summary>  ### Visual Description ## Diagram: LLM Agent Interaction with Environment and Memory ### Overview The image is a diagram illustrating the interaction of LLM (Large Language Model) Agents with an environment and an agentic memory. It shows a cyclical flow of information between these three components. ### Components/Axes * **Environment:** Represented by a globe icon with blue oceans and green landmasses. * **LLM Agents:** Represented by a blue robot-like icon with a speech bubble. * **Agentic Memory:** Represented by a blue robot-like icon with a speech bubble, connected to three horizontal memory blocks. * **Interaction:** A bidirectional arrow connecting the Environment and LLM Agents, labeled "Interaction". * **Write:** A unidirectional arrow pointing from LLM Agents to Agentic Memory, labeled "Write". * **Read:** A unidirectional arrow pointing from Agentic Memory to LLM Agents, labeled "Read". ### Detailed Analysis * The "Environment" interacts with the "LLM Agents" through a two-way "Interaction". * The "LLM Agents" "Write" to the "Agentic Memory". * The "Agentic Memory" is read by the "LLM Agents" through a "Read" operation. * The "Agentic Memory" consists of three horizontal blocks, each with an arrow pointing towards the "LLM Agents", indicating the flow of information during the "Read" operation. ### Key Observations * The diagram illustrates a closed-loop system where LLM Agents interact with the environment, store information in memory, and retrieve information from memory. * The use of robot-like icons for LLM Agents and Agentic Memory suggests that these components are intelligent and capable of processing information. ### Interpretation The diagram depicts a system where LLM Agents learn and adapt through interaction with their environment and by utilizing a memory component. The "Interaction" with the environment provides the agents with new information, which they can then "Write" to their "Agentic Memory". This stored information can then be "Read" by the agents to inform their future interactions with the environment. This cycle allows the agents to improve their performance over time by learning from their experiences. The three memory blocks suggest that the agentic memory may be structured or tiered in some way. </details> (b) Our proposed agentic memory. Figure 1: Traditional memory systems require predefined memory access patterns specified in the workflow, limiting their adaptability to diverse scenarios. Contrastly, our A-Mem enhances the flexibility of LLM agents by enabling dynamic memory operations. In this paper, we introduce a novel agentic memory system, named as A-Mem, for LLM agents that enables dynamic memory structuring without relying on static, predetermined memory operations. Our approach draws inspiration from the Zettelkasten method [15, 1], a sophisticated knowledge management system that creates interconnected information networks through atomic notes and flexible linking mechanisms. Our system introduces an agentic memory architecture that enables autonomous and flexible memory management for LLM agents. For each new memory, we construct comprehensive notes, which integrates multiple representations: structured textual attributes including several attributes and embedding vectors for similarity matching. Then A-Mem analyzes the historical memory repository to establish meaningful connections based on semantic similarities and shared attributes. This integration process not only creates new links but also enables dynamic evolution when new memories are incorporated, they can trigger updates to the contextual representations of existing memories, allowing the entire memories to continuously refine and deepen its understanding over time. The contributions are summarized as: We present A-Mem, an agentic memory system for LLM agents that enables autonomous generation of contextual descriptions, dynamic establishment of memory connections, and intelligent evolution of existing memories based on new experiences. This system equips LLM agents with long-term interaction capabilities without requiring predetermined memory operations. We design an agentic memory update mechanism where new memories automatically trigger two key operations: link generation and memory evolution. Link generation automatically establishes connections between memories by identifying shared attributes and similar contextual descriptions. Memory evolution enables existing memories to dynamically adapt as new experiences are analyzed, leading to the emergence of higher-order patterns and attributes. We conduct comprehensive evaluations of our system using a long-term conversational dataset, comparing performance across six foundation models using six distinct evaluation metrics, demonstrating significant improvements. Moreover, we provide T-SNE visualizations to illustrate the structured organization of our agentic memory system. 2 Related Work 2.1 Memory for LLM Agents Prior works on LLM agent memory systems have explored various mechanisms for memory management and utilization [23, 21, 8, 39]. Some approaches complete interaction storage, which maintains comprehensive historical records through dense retrieval models [39] or read-write memory structures [24]. Moreover, MemGPT [25] leverages cache-like architectures to prioritize recent information. Similarly, SCM [32] proposes a Self-Controlled Memory framework that enhances LLMs’ capability to maintain long-term memory through a memory stream and controller mechanism. However, these approaches face significant limitations in handling diverse real-world tasks. While they can provide basic memory functionality, their operations are typically constrained by predefined structures and fixed workflows. These constraints stem from their reliance on rigid operational patterns, particularly in memory writing and retrieval processes. Such inflexibility leads to poor generalization in new environments and limited effectiveness in long-term interactions. Therefore, designing a flexible and universal memory system that supports agents’ long-term interactions remains a crucial challenge. 2.2 Retrieval-Augmented Generation Retrieval-Augmented Generation (RAG) has emerged as a powerful approach to enhance LLMs by incorporating external knowledge sources [18, 6, 10]. The standard RAG [37, 34] process involves indexing documents into chunks, retrieving relevant chunks based on semantic similarity, and augmenting the LLM’s prompt with this retrieved context for generation. Advanced RAG systems [20, 12] have evolved to include sophisticated pre-retrieval and post-retrieval optimizations. Building upon these foundations, recent researches has introduced agentic RAG systems that demonstrate more autonomous and adaptive behaviors in the retrieval process. These systems can dynamically determine when and what to retrieve [4, 14], generate hypothetical responses to guide retrieval, and iteratively refine their search strategies based on intermediate results [31, 29]. However, while agentic RAG approaches demonstrate agency in the retrieval phase by autonomously deciding when and what to retrieve [4, 14, 38], our agentic memory system exhibits agency at a more fundamental level through the autonomous evolution of its memory structure. Inspired by the Zettelkasten method, our system allows memories to actively generate their own contextual descriptions, form meaningful connections with related memories, and evolve both their content and relationships as new experiences emerge. This fundamental distinction in agency between retrieval versus storage and evolution distinguishes our approach from agentic RAG systems, which maintain static knowledge bases despite their sophisticated retrieval mechanisms. 3 Methodolodgy Our proposed agentic memory system draws inspiration from the Zettelkasten method, implementing a dynamic and self-evolving memory system that enables LLM agents to maintain long-term memory without predetermined operations. The system’s design emphasizes atomic note-taking, flexible linking mechanisms, and continuous evolution of knowledge structures. <details> <summary>x5.png Details</summary>  ### Visual Description ## Diagram: LLM Memory Management ### Overview The image is a diagram illustrating the process of memory management using Large Language Models (LLMs). It is divided into four main sections: Note Construction, Link Generation, Memory Evolution, and Memory Retrieval. The diagram shows how user interactions are converted into notes, linked in memory, evolved over time, and retrieved when needed. ### Components/Axes * **Titles:** * Note Construction (Top-Left) * Link Generation (Top-Center-Left) * Memory Evolution (Top-Center-Right) * Memory Retrieval (Top-Right) * **Note Construction:** * Environment: Depicted as a globe. * LLM Agents: Depicted as a blue chatbot icon. * Interaction: A bidirectional arrow between the Environment and LLM Agents. * Write: A downward arrow indicating the writing process. * Conversation 1: A text box containing the question, "Can you help me implement a custom cache system for my web application? I need it to handle both memory and disk storage." * Conversation 2: A text box containing the statement, "The cache system works great, but we're seeing high memory usage in production. Can we modify it to implement an LRU eviction policy?" * LLM: Represented by the LLM logo. * Note Attributes: * Timestamp: Clock icon. * Content: List icon. * Context: Document icon. * Keywords: Tag icon. * Tags: Orange tag icon. * Embedding: A series of small rectangles. * Note: Represented as a blue notebook icon. * **Link Generation:** * Memory: Represented as a database icon. * Boxes: Labeled as Box 1, Box i, Box j, Box n. Each box contains three blue notebook icons. * Top-k: A downward arrow indicating the top-k selection process. * m: A node with three blue notebook icons, connected to other nodes via lines and small circular icons. * Store: A downward arrow indicating the storing process. * Boxes: Labeled as Box n+1, Box n+2. Each box contains three blue notebook icons. * **Memory Evolution:** * Boxes: Labeled as Box n+1, Box n+2. Each box contains three blue notebook icons. * LLM: Represented by the LLM logo. * Action: A downward arrow indicating the action process. * Evolve: A node with three blue notebook icons and a gear icon. * **Memory Retrieval:** * Retrieve: An arrow pointing from the "Memory Evolution" section to a "Query Embedding" box. * Query Embedding: A horizontal box labeled "Query Embedding" with "Text Model" above it. * Query: A question mark icon. * Top-k: An arrow pointing from the "Query Embedding" box to the "Relative Memory" box. * Relative Memory: A box containing three blue notebook icons, with the leftmost icon highlighted with a pink circle and labeled "1st". * LLM Agents: Depicted as a blue chatbot icon. ### Detailed Analysis or Content Details * **Note Construction:** The process starts with an interaction between the environment and LLM agents. The agents write conversations, which are then processed by LLMs to create notes. These notes have attributes such as timestamp, content, context, keywords, tags, and embeddings. * **Link Generation:** The notes are stored in memory boxes (Box 1 to Box n). A top-k selection process retrieves relevant notes, which are then stored in new memory boxes (Box n+1 and Box n+2). * **Memory Evolution:** The notes in memory boxes (Box n+1 and Box n+2) are processed by an LLM, which takes action and evolves the notes. * **Memory Retrieval:** A query is embedded using a text model. A top-k selection process retrieves relative memory, and the first (1st) item is highlighted. The retrieved information is then used by LLM agents. ### Key Observations * The diagram illustrates a cyclical process of note creation, storage, evolution, and retrieval. * LLMs play a central role in processing and evolving the notes. * The top-k selection process is used in both link generation and memory retrieval. * The diagram highlights the importance of note attributes in the memory management process. ### Interpretation The diagram presents a high-level overview of how LLMs can be used for memory management. The process involves converting user interactions into structured notes, linking these notes in memory, evolving them over time, and retrieving them when needed. The use of top-k selection suggests a mechanism for prioritizing relevant information. The diagram emphasizes the importance of context and attributes in managing and retrieving information effectively. The cyclical nature of the process indicates a continuous learning and adaptation mechanism. </details> Figure 2: Our A-Mem architecture comprises three integral parts in memory storage. During note construction, the system processes new interaction memories and stores them as notes with multiple attributes. The link generation process first retrieves the most relevant historical memories and then employs an LLM to determine whether connections should be established between them. The concept of a ’box’ describes that related memories become interconnected through their similar contextual descriptions, analogous to the Zettelkasten method. However, our approach allows individual memories to exist simultaneously within multiple different boxes. During the memory retrieval stage, we extract query embeddings using a text encoding model and search the memory database for relevant matches. When related memory is retrieved, similar memories that are linked within the same box are also automatically accessed. 3.1 Note Construction Building upon the Zettelkasten method’s principles of atomic note-taking and flexible organization, we introduce an LLM-driven approach to memory note construction. When an agent interacts with its environment, we construct structured memory notes that capture both explicit information and LLM-generated contextual understanding. Each memory note $m_{i}$ in our collection $\mathcal{M}=\{m_{1},m_{2},...,m_{N}\}$ is represented as: $$ m_{i}=\{c_{i},t_{i},K_{i},G_{i},X_{i},e_{i},L_{i}\} \tag{1} $$ where $c_{i}$ represents the original interaction content, $t_{i}$ is the timestamp of the interaction, $K_{i}$ denotes LLM-generated keywords that capture key concepts, $G_{i}$ contains LLM-generated tags for categorization, $X_{i}$ represents the LLM-generated contextual description that provides rich semantic understanding, and $L_{i}$ maintains the set of linked memories that share semantic relationships. To enrich each memory note with meaningful context beyond its basic content and timestamp, we leverage an LLM to analyze the interaction and generate these semantic components. The note construction process involves prompting the LLM with carefully designed templates $P_{s1}$ : $$ K_{i},G_{i},X_{i}\leftarrow\text{LLM}(c_{i}\;\Vert t_{i}\;\Vert P_{s1}) \tag{2} $$ Following the Zettelkasten principle of atomicity, each note captures a single, self-contained unit of knowledge. To enable efficient retrieval and linking, we compute a dense vector representation via a text encoder [27] that encapsulates all textual components of the note: $$ e_{i}=f_{\text{enc}}[\;\text{concat}(c_{i},K_{i},G_{i},X_{i})\;] \tag{3} $$ By using LLMs to generate enriched components, we enable autonomous extraction of implicit knowledge from raw interactions. The multi-faceted note structure ( $K_{i}$ , $G_{i}$ , $X_{i}$ ) creates rich representations that capture different aspects of the memory, facilitating nuanced organization and retrieval. Additionally, the combination of LLM-generated semantic components with dense vector representations provides both context and computationally efficient similarity matching. 3.2 Link Generation Our system implements an autonomous link generation mechanism that enables new memory notes to form meaningful connections without predefined rules. When the constrctd memory note $m_{n}$ is added to the system, we first leverage its semantic embedding for similarity-based retrieval. For each existing memory note $m_{j}∈\mathcal{M}$ , we compute a similarity score: $$ s_{n,j}=\frac{e_{n}\cdot e_{j}}{|e_{n}||e_{j}|} \tag{4} $$ The system then identifies the top- $k$ most relevant memories: $$ \mathcal{M}_{\text{near}}^{n}=\{m_{j}|\;\text{rank}(s_{n,j})\leq k,m_{j}\in\mathcal{M}\} \tag{5} $$ Based on these candidate nearest memories, we prompt the LLM to analyze potential connections based on their potential common attributes. Formally, the link set of memory $m_{n}$ update like: $$ L_{i}\leftarrow\text{LLM}(m_{n}\;\Vert\mathcal{M}_{\text{near}}^{n}\;\Vert P_{s2}) \tag{6} $$ Each generated link $l_{i}$ is structured as: $L_{i}=\{m_{i},...,m_{k}\}$ . By using embedding-based retrieval as an initial filter, we enable efficient scalability while maintaining semantic relevance. A-Mem can quickly identify potential connections even in large memory collections without exhaustive comparison. More importantly, the LLM-driven analysis allows for nuanced understanding of relationships that goes beyond simple similarity metrics. The language model can identify subtle patterns, causal relationships, and conceptual connections that might not be apparent from embedding similarity alone. We implements the Zettelkasten principle of flexible linking while leveraging modern language models. The resulting network emerges organically from memory content and context, enabling natural knowledge organization. 3.3 Memory Evolution After creating links for the new memory, A-Mem evolves the retrieved memories based on their textual information and relationships with the new memory. For each memory $m_{j}$ in the nearest neighbor set $\mathcal{M}_{\text{near}}^{n}$ , the system determines whether to update its context, keywords, and tags. This evolution process can be formally expressed as: $$ m_{j}^{*}\leftarrow\text{LLM}(m_{n}\;\Vert\mathcal{M}_{\text{near}}^{n}\setminus m_{j}\;\Vert m_{j}\;\Vert P_{s3}) \tag{7} $$ The evolved memory $m_{j}^{*}$ then replaces the original memory $m_{j}$ in the memory set $\mathcal{M}$ . This evolutionary approach enables continuous updates and new connections, mimicking human learning processes. As the system processes more memories over time, it develops increasingly sophisticated knowledge structures, discovering higher-order patterns and concepts across multiple memories. This creates a foundation for autonomous memory learning where knowledge organization becomes progressively richer through the ongoing interaction between new experiences and existing memories. 3.4 Retrieve Relative Memory In each interaction, our A-Mem performs context-aware memory retrieval to provide the agent with relevant historical information. Given a query text $q$ from the current interaction, we first compute its dense vector representation using the same text encoder used for memory notes: $$ e_{q}=f_{\text{enc}}(q) \tag{8} $$ The system then computes similarity scores between the query embedding and all existing memory notes in $\mathcal{M}$ using cosine similarity: $$ s_{q,i}=\frac{e_{q}\cdot e_{i}}{|e_{q}||e_{i}|},\text{where}\;e_{i}\in m_{i},\;\forall m_{i}\in\mathcal{M} \tag{9} $$ Then we retrieve the k most relevant memories from the historical memory storage to construct a contextually appropriate prompt. $$ \mathcal{M}_{\text{retrieved}}=\{m_{i}|\text{rank}(s_{q,i})\leq k,m_{i}\in\mathcal{M}\} \tag{10} $$ These retrieved memories provide relevant historical context that helps the agent better understand and respond to the current interaction. The retrieved context enriches the agent’s reasoning process by connecting the current interaction with related past experiences stored in the memory system. 4 Experiment 4.1 Dataset and Evaluation To evaluate the effectiveness of instruction-aware recommendation in long-term conversations, we utilize the LoCoMo dataset [22], which contains significantly longer dialogues compared to existing conversational datasets [36, 13]. While previous datasets contain dialogues with around 1K tokens over 4-5 sessions, LoCoMo features much longer conversations averaging 9K tokens spanning up to 35 sessions, making it particularly suitable for evaluating models’ ability to handle long-range dependencies and maintain consistency over extended conversations. The LoCoMo dataset comprises diverse question types designed to comprehensively evaluate different aspects of model understanding: (1) single-hop questions answerable from a single session; (2) multi-hop questions requiring information synthesis across sessions; (3) temporal reasoning questions testing understanding of time-related information; (4) open-domain knowledge questions requiring integration of conversation context with external knowledge; and (5) adversarial questions assessing models’ ability to identify unanswerable queries. In total, LoCoMo contains 7,512 question-answer pairs across these categories. Besides, we use a new dataset, named DialSim [16], to evaluate the effectiveness of our memory system. It is question-answering dataset derived from long-term multi-party dialogues. The dataset is derived from popular TV shows (Friends, The Big Bang Theory, and The Office), covering 1,300 sessions spanning five years, containing approximately 350,000 tokens, and including more than 1,000 questions per session from refined fan quiz website questions and complex questions generated from temporal knowledge graphs. For comparison baselines, we compare to LoCoMo [22], ReadAgent [17], MemoryBank [39] and MemGPT [25]. The detailed introduction of baselines can be found in Appendix A.1 For evaluation, we employ two primary metrics: the F1 score to assess answer accuracy by balancing precision and recall, and BLEU-1 [26] to evaluate generated response quality by measuring word overlap with ground truth responses. Also, we report the average token length for answering one question. Besides reporting experiment results with four additional metrics (ROUGE-L, ROUGE-2, METEOR, and SBERT Similarity), we also present experimental outcomes using different foundation models including DeepSeek-R1-32B [11], Claude 3.0 Haiku [2], and Claude 3.5 Haiku [3] in Appendix A.3. 4.2 Implementation Details For all baselines and our proposed method, we maintain consistency by employing identical system prompts as detailed in Appendix B. The deployment of Qwen-1.5B/3B and Llama 3.2 1B/3B models is accomplished through local instantiation using Ollama https://github.com/ollama/ollama, with LiteLLM https://github.com/BerriAI/litellm managing structured output generation. For GPT models, we utilize the official structured output API. In our memory retrieval process, we primarily employ $k$ =10 for top- $k$ memory selection to maintain computational efficiency, while adjusting this parameter for specific categories to optimize performance. The detailed configurations of $k$ can be found in Appendix A.5. For text embedding, we implement the all-minilm-l6-v2 model across all experiments. Table 1: Experimental results on LoCoMo dataset of QA tasks across five categories (Multi Hop, Temporal, Open Domain, Single Hop, and Adversial) using different methods. Results are reported in F1 and BLEU-1 (%) scores. The best performance is marked in bold, and our proposed method A-Mem (highlighted in gray) demonstrates competitive performance across six foundation language models. | Model | Method | Category | Average | | | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | Ranking | Token | | | | | | | | | | | F1 | BLEU | F1 | BLEU | F1 | BLEU | F1 | BLEU | F1 | BLEU | F1 | BLEU | Length | | | | | GPT | 4o-mini | LoCoMo | 25.02 | 19.75 | 18.41 | 14.77 | 12.04 | 11.16 | 40.36 | 29.05 | 69.23 | 68.75 | 2.4 | 2.4 | 16,910 | | ReadAgent | 9.15 | 6.48 | 12.60 | 8.87 | 5.31 | 5.12 | 9.67 | 7.66 | 9.81 | 9.02 | 4.2 | 4.2 | 643 | | | | MemoryBank | 5.00 | 4.77 | 9.68 | 6.99 | 5.56 | 5.94 | 6.61 | 5.16 | 7.36 | 6.48 | 4.8 | 4.8 | 432 | | | | MemGPT | 26.65 | 17.72 | 25.52 | 19.44 | 9.15 | 7.44 | 41.04 | 34.34 | 43.29 | 42.73 | 2.4 | 2.4 | 16,977 | | | | A-Mem | 27.02 | 20.09 | 45.85 | 36.67 | 12.14 | 12.00 | 44.65 | 37.06 | 50.03 | 49.47 | 1.2 | 1.2 | 2,520 | | | | 4o | LoCoMo | 28.00 | 18.47 | 9.09 | 5.78 | 16.47 | 14.80 | 61.56 | 54.19 | 52.61 | 51.13 | 2.0 | 2.0 | 16,910 | | | ReadAgent | 14.61 | 9.95 | 4.16 | 3.19 | 8.84 | 8.37 | 12.46 | 10.29 | 6.81 | 6.13 | 4.0 | 4.0 | 805 | | | | MemoryBank | 6.49 | 4.69 | 2.47 | 2.43 | 6.43 | 5.30 | 8.28 | 7.10 | 4.42 | 3.67 | 5.0 | 5.0 | 569 | | | | MemGPT | 30.36 | 22.83 | 17.29 | 13.18 | 12.24 | 11.87 | 60.16 | 53.35 | 34.96 | 34.25 | 2.4 | 2.4 | 16,987 | | | | A-Mem | 32.86 | 23.76 | 39.41 | 31.23 | 17.10 | 15.84 | 48.43 | 42.97 | 36.35 | 35.53 | 1.6 | 1.6 | 1,216 | | | | Qwen2.5 | 1.5b | LoCoMo | 9.05 | 6.55 | 4.25 | 4.04 | 9.91 | 8.50 | 11.15 | 8.67 | 40.38 | 40.23 | 3.4 | 3.4 | 16,910 | | ReadAgent | 6.61 | 4.93 | 2.55 | 2.51 | 5.31 | 12.24 | 10.13 | 7.54 | 5.42 | 27.32 | 4.6 | 4.6 | 752 | | | | MemoryBank | 11.14 | 8.25 | 4.46 | 2.87 | 8.05 | 6.21 | 13.42 | 11.01 | 36.76 | 34.00 | 2.6 | 2.6 | 284 | | | | MemGPT | 10.44 | 7.61 | 4.21 | 3.89 | 13.42 | 11.64 | 9.56 | 7.34 | 31.51 | 28.90 | 3.4 | 3.4 | 16,953 | | | | A-Mem | 18.23 | 11.94 | 24.32 | 19.74 | 16.48 | 14.31 | 23.63 | 19.23 | 46.00 | 43.26 | 1.0 | 1.0 | 1,300 | | | | 3b | LoCoMo | 4.61 | 4.29 | 3.11 | 2.71 | 4.55 | 5.97 | 7.03 | 5.69 | 16.95 | 14.81 | 3.2 | 3.2 | 16,910 | | | ReadAgent | 2.47 | 1.78 | 3.01 | 3.01 | 5.57 | 5.22 | 3.25 | 2.51 | 15.78 | 14.01 | 4.2 | 4.2 | 776 | | | | MemoryBank | 3.60 | 3.39 | 1.72 | 1.97 | 6.63 | 6.58 | 4.11 | 3.32 | 13.07 | 10.30 | 4.2 | 4.2 | 298 | | | | MemGPT | 5.07 | 4.31 | 2.94 | 2.95 | 7.04 | 7.10 | 7.26 | 5.52 | 14.47 | 12.39 | 2.4 | 2.4 | 16,961 | | | | A-Mem | 12.57 | 9.01 | 27.59 | 25.07 | 7.12 | 7.28 | 17.23 | 13.12 | 27.91 | 25.15 | 1.0 | 1.0 | 1,137 | | | | Llama 3.2 | 1b | LoCoMo | 11.25 | 9.18 | 7.38 | 6.82 | 11.90 | 10.38 | 12.86 | 10.50 | 51.89 | 48.27 | 3.4 | 3.4 | 16,910 | | ReadAgent | 5.96 | 5.12 | 1.93 | 2.30 | 12.46 | 11.17 | 7.75 | 6.03 | 44.64 | 40.15 | 4.6 | 4.6 | 665 | | | | MemoryBank | 13.18 | 10.03 | 7.61 | 6.27 | 15.78 | 12.94 | 17.30 | 14.03 | 52.61 | 47.53 | 2.0 | 2.0 | 274 | | | | MemGPT | 9.19 | 6.96 | 4.02 | 4.79 | 11.14 | 8.24 | 10.16 | 7.68 | 49.75 | 45.11 | 4.0 | 4.0 | 16,950 | | | | A-Mem | 19.06 | 11.71 | 17.80 | 10.28 | 17.55 | 14.67 | 28.51 | 24.13 | 58.81 | 54.28 | 1.0 | 1.0 | 1,376 | | | | 3b | LoCoMo | 6.88 | 5.77 | 4.37 | 4.40 | 10.65 | 9.29 | 8.37 | 6.93 | 30.25 | 28.46 | 2.8 | 2.8 | 16,910 | | | ReadAgent | 2.47 | 1.78 | 3.01 | 3.01 | 5.57 | 5.22 | 3.25 | 2.51 | 15.78 | 14.01 | 4.2 | 4.2 | 461 | | | | MemoryBank | 6.19 | 4.47 | 3.49 | 3.13 | 4.07 | 4.57 | 7.61 | 6.03 | 18.65 | 17.05 | 3.2 | 3.2 | 263 | | | | MemGPT | 5.32 | 3.99 | 2.68 | 2.72 | 5.64 | 5.54 | 4.32 | 3.51 | 21.45 | 19.37 | 3.8 | 3.8 | 16,956 | | | | A-Mem | 17.44 | 11.74 | 26.38 | 19.50 | 12.53 | 11.83 | 28.14 | 23.87 | 42.04 | 40.60 | 1.0 | 1.0 | 1,126 | | | 4.3 Empricial Results Performance Analysis. In our empirical evaluation, we compared A-Mem with four competitive baselines including LoCoMo [22], ReadAgent [17], MemoryBank [39], and MemGPT [25] on the LoCoMo dataset. For non-GPT foundation models, our A-Mem consistently outperforms all baselines across different categories, demonstrating the effectiveness of our agentic memory approach. For GPT-based models, while LoCoMo and MemGPT show strong performance in certain categories like Open Domain and Adversial tasks due to their robust pre-trained knowledge in simple fact retrieval, our A-Mem demonstrates superior performance in Multi-Hop tasks achieves at least two times better performance that require complex reasoning chains. In addition to experiments on the LoCoMo dataset, we also compare our method on the DialSim dataset against LoCoMo and MemGPT. A-Mem consistently outperforms all baselines across evaluation metrics, achieving an F1 score of 3.45 (a 35% improvement over LoCoMo’s 2.55 and 192% higher than MemGPT’s 1.18). The effectiveness of A-Mem stems from its novel agentic memory architecture that enables dynamic and structured memory management. Unlike traditional approaches that use static memory operations, our system creates interconnected memory networks through atomic notes with rich contextual descriptions, enabling more effective multi-hop reasoning. The system’s ability to dynamically establish connections between memories based on shared attributes and continuously update existing memory descriptions with new contextual information allows it to better capture and utilize the relationships between different pieces of information. Table 2: Comparison of different memory mechanisms across multiple evaluation metrics on DialSim [16]. Higher scores indicate better performance, with A-Mem showing superior results across all metrics. | Method | F1 | BLEU-1 | ROUGE-L | ROUGE-2 | METEOR | SBERT Similarity | | --- | --- | --- | --- | --- | --- | --- | | LoCoMo | 2.55 | 3.13 | 2.75 | 0.90 | 1.64 | 15.76 | | MemGPT | 1.18 | 1.07 | 0.96 | 0.42 | 0.95 | 8.54 | | A-Mem | 3.45 | 3.37 | 3.54 | 3.60 | 2.05 | 19.51 | Cost-Efficiency Analysis. A-Mem demonstrates significant computational and cost efficiency alongside strong performance. The system requires approximately 1,200 tokens per memory operation, achieving an 85-93% reduction in token usage compared to baseline methods (LoCoMo and MemGPT with 16,900 tokens) through our selective top-k retrieval mechanism. This substantial token reduction directly translates to lower operational costs, with each memory operation costing less than $0.0003 when using commercial API services—making large-scale deployments economically viable. Processing times average 5.4 seconds using GPT-4o-mini and only 1.1 seconds with locally-hosted Llama 3.2 1B on a single GPU. Despite requiring multiple LLM calls during memory processing, A-Mem maintains this cost-effective resource utilization while consistently outperforming baseline approaches across all foundation models tested, particularly doubling performance on complex multi-hop reasoning tasks. This balance of low computational cost and superior reasoning capability highlights A-Mem ’s practical advantage for deployment in the real world. Table 3: An ablation study was conducted to evaluate our proposed method against the GPT-4o-mini base model. The notation ’w/o’ indicates experiments where specific modules were removed. The abbreviations LG and ME denote the link generation module and memory evolution module, respectively. | Method | Category | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | | | | | | | | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | | | w/o LG & ME | 9.65 | 7.09 | 24.55 | 19.48 | 7.77 | 6.70 | 13.28 | 10.30 | 15.32 | 18.02 | | w/o ME | 21.35 | 15.13 | 31.24 | 27.31 | 10.13 | 10.85 | 39.17 | 34.70 | 44.16 | 45.33 | | A-Mem | 27.02 | 20.09 | 45.85 | 36.67 | 12.14 | 12.00 | 44.65 | 37.06 | 50.03 | 49.47 | 4.4 Ablation Study To evaluate the effectiveness of the Link Generation (LG) and Memory Evolution (ME) modules, we conduct the ablation study by systematically removing key components of our model. When both LG and ME modules are removed, the system exhibits substantial performance degradation, particularly in Multi Hop reasoning and Open Domain tasks. The system with only LG active (w/o ME) shows intermediate performance levels, maintaining significantly better results than the version without both modules, which demonstrates the fundamental importance of link generation in establishing memory connections. Our full model, A-Mem, consistently achieves the best performance across all evaluation categories, with particularly strong results in complex reasoning tasks. These results reveal that while the link generation module serves as a critical foundation for memory organization, the memory evolution module provides essential refinements to the memory structure. The ablation study validates our architectural design choices and highlights the complementary nature of these two modules in creating an effective memory system. <details> <summary>x6.png Details</summary>  ### Visual Description ## Bar Chart: F1 and BLEU Scores vs. k Values ### Overview The image is a bar chart comparing F1 and BLEU scores for different values of 'k'. The x-axis represents 'k values' ranging from 10 to 50, while the y-axis represents the scores, ranging from 12.5 to 27.5. Two data series are displayed: F1 (blue) and BLEU (orange). ### Components/Axes * **Title:** There is no explicit title. * **X-axis:** 'k values' with markers at 10, 20, 30, 40, and 50. * **Y-axis:** Numerical scale ranging from 12.5 to 27.5, with increments of 2.5. * **Legend:** Located in the top-left corner. * Blue: F1 * Orange: BLEU * **Gridlines:** Horizontal gridlines are present to aid in value estimation. ### Detailed Analysis The chart presents F1 and BLEU scores for k values of 10, 20, 30, 40, and 50. * **F1 (Blue):** * k=10: 19.91 * k=20: 25.87 * k=30: 26.97 * k=40: 27.02 * k=50: 26.81 The F1 score increases significantly from k=10 to k=20, then shows a slight increase to k=40, followed by a minor decrease at k=50. * **BLEU (Orange):** * k=10: 14.36 * k=20: 19.45 * k=30: 20.19 * k=40: 20.09 * k=50: 20.15 The BLEU score increases from k=10 to k=30, then plateaus with minor fluctuations between k=30 and k=50. ### Key Observations * The F1 score is consistently higher than the BLEU score for all k values. * The F1 score shows a more pronounced increase with increasing k values compared to the BLEU score. * The BLEU score plateaus after k=30, suggesting diminishing returns for higher k values. ### Interpretation The chart suggests that increasing the 'k value' initially improves both F1 and BLEU scores. However, the BLEU score plateaus after k=30, while the F1 score reaches its peak at k=40. This indicates that there is an optimal 'k value' for maximizing the F1 score, beyond which further increases in 'k' do not significantly improve performance and may even lead to a slight decrease. The F1 score appears to be more sensitive to changes in 'k' compared to the BLEU score. The data suggests that a 'k value' around 30-40 might be optimal for this particular scenario, balancing both F1 and BLEU scores. </details> (a) Multi Hop <details> <summary>x7.png Details</summary>  ### Visual Description ## Bar Chart: F1 and BLEU-13 Scores vs. k Values ### Overview The image is a bar chart comparing F1 and BLEU-13 scores for different values of 'k'. The chart displays two sets of vertical bars for each 'k' value, one representing the F1 score (blue) and the other representing the BLEU-13 score (orange). The x-axis represents 'k values', and the y-axis represents the scores. ### Components/Axes * **Title:** There is no explicit title on the chart. * **X-axis:** * Label: "k values" * Values: 10, 20, 30, 40, 50 * **Y-axis:** * Scale: 35.0 to 47.5, incrementing by 2.5 * **Legend:** Located in the top-left corner. * Blue bar: "F1" * Orange bar: "BLEU-13" * **Grid:** The chart has a light gray grid in the background. ### Detailed Analysis The chart presents F1 and BLEU-13 scores for k values of 10, 20, 30, 40, and 50. **F1 Scores (Blue Bars):** * k = 10: 43.60 * k = 20: 45.03 * k = 30: 45.22 * k = 40: 45.85 * k = 50: 45.60 Trend: The F1 score generally increases from k=10 to k=40, then slightly decreases at k=50. **BLEU-13 Scores (Orange Bars):** * k = 10: 35.53 * k = 20: 35.85 * k = 30: 36.44 * k = 40: 36.67 * k = 50: 35.76 Trend: The BLEU-13 score increases from k=10 to k=40, then decreases at k=50. ### Key Observations * The F1 scores are consistently higher than the BLEU-13 scores for all 'k' values. * Both F1 and BLEU-13 scores peak around k=40. * The F1 score shows a more pronounced increase from k=10 to k=20 compared to the other intervals. ### Interpretation The chart suggests that increasing the 'k' value initially improves both F1 and BLEU-13 scores, indicating better performance up to a certain point. The slight decrease in scores at k=50 suggests that there might be a point of diminishing returns or even a negative impact on performance with excessively high 'k' values. The F1 score, being consistently higher, might be a more robust metric in this context compared to BLEU-13. The optimal 'k' value appears to be around 40, where both metrics achieve relatively high scores. </details> (b) Temporal <details> <summary>x8.png Details</summary>  ### Visual Description ## Bar Chart: F1 and BLEU-1 Scores vs. k Values ### Overview The image is a bar chart comparing F1 and BLEU-1 scores for different values of 'k'. The x-axis represents 'k values' ranging from 10 to 50, and the y-axis represents the score. Each 'k value' has two bars, one for F1 score (blue) and one for BLEU-1 score (orange). Numerical values are displayed above each bar. ### Components/Axes * **X-axis:** 'k values' with markers at 10, 20, 30, 40, and 50. * **Y-axis:** Score, ranging from 6 to 14. No explicit units are given. * **Legend:** Located in the top-left corner. * Blue bar: F1 * Orange bar: BLEU-1 * **Gridlines:** Horizontal gridlines are present at intervals of 2, starting from 6. ### Detailed Analysis The chart presents F1 and BLEU-1 scores for k values of 10, 20, 30, 40, and 50. * **k = 10:** * F1 (blue): 7.38 * BLEU-1 (orange): 7.03 * **k = 20:** * F1 (blue): 10.29 * BLEU-1 (orange): 9.61 * **k = 30:** * F1 (blue): 12.24 * BLEU-1 (orange): 10.57 * **k = 40:** * F1 (blue): 10.35 * BLEU-1 (orange): 9.76 * **k = 50:** * F1 (blue): 12.11 * BLEU-1 (orange): 12.00 **Trends:** * **F1 Score:** The F1 score generally increases from k=10 to k=30, then decreases at k=40, and increases again at k=50. * **BLEU-1 Score:** The BLEU-1 score generally increases from k=10 to k=30, then decreases at k=40, and increases again at k=50. ### Key Observations * For all k values, the F1 score is higher than the BLEU-1 score, except at k=50 where the BLEU-1 score is slightly lower. * The highest F1 score is observed at k=30 (12.24). * The highest BLEU-1 score is observed at k=50 (12.00). * Both F1 and BLEU-1 scores show a similar trend, increasing initially, then decreasing, and increasing again. ### Interpretation The chart compares the performance of a system using F1 and BLEU-1 metrics for different values of 'k'. The data suggests that increasing 'k' initially improves both F1 and BLEU-1 scores, but there's a point (around k=40) where performance dips before recovering at k=50. The optimal 'k' value, based on this data, appears to be around 30 for F1 and 50 for BLEU-1. The relationship between F1 and BLEU-1 is generally consistent, with F1 scores being slightly higher than BLEU-1 scores for most 'k' values. This could indicate that the system is better at balancing precision and recall (F1) than at matching n-grams (BLEU-1), except at k=50 where they are nearly equal. </details> (c) Open Domain <details> <summary>x9.png Details</summary>  ### Visual Description ## Bar Chart: F1 and BLEU-1 Scores vs. k Values ### Overview The image is a bar chart comparing F1 and BLEU-1 scores for different values of 'k'. The x-axis represents 'k values' ranging from 10 to 50, and the y-axis represents the scores. The chart displays two data series: F1 scores (represented by blue bars) and BLEU-1 scores (represented by orange bars). ### Components/Axes * **X-axis:** 'k values' with markers at 10, 20, 30, 40, and 50. * **Y-axis:** Numerical scale ranging from approximately 25 to 45, with gridlines at intervals of 5. * **Legend (top-left):** * Blue: F1 * Orange: BLEU-1 ### Detailed Analysis * **F1 (Blue Bars):** The F1 score generally increases as the 'k value' increases. * k = 10: F1 = 31.15 * k = 20: F1 = 33.67 * k = 30: F1 = 38.15 * k = 40: F1 = 41.55 * k = 50: F1 = 44.55 * **BLEU-1 (Orange Bars):** The BLEU-1 score also increases as the 'k value' increases. * k = 10: BLEU-1 = 25.43 * k = 20: BLEU-1 = 28.31 * k = 30: BLEU-1 = 32.12 * k = 40: BLEU-1 = 34.32 * k = 50: BLEU-1 = 37.02 ### Key Observations * The F1 score is consistently higher than the BLEU-1 score for all 'k values'. * Both F1 and BLEU-1 scores show a positive correlation with 'k values'. * The increase in F1 score appears to be more pronounced than the increase in BLEU-1 score as 'k' increases. ### Interpretation The chart suggests that increasing the 'k value' improves both F1 and BLEU-1 scores, indicating better performance in whatever task these metrics are evaluating. The F1 score consistently outperforms the BLEU-1 score, implying that the system or model being evaluated performs better according to the F1 metric. The increasing trend suggests that further increasing 'k' might lead to even higher scores, although this is not explicitly shown in the chart. The relationship between 'k' and these metrics is likely related to the specific algorithm or model being used, and further investigation would be needed to understand the underlying reasons for this trend. </details> (d) Single Hop <details> <summary>x10.png Details</summary>  ### Visual Description ## Bar Chart: F1 and BLEU-1 Scores vs. k Values ### Overview The image is a bar chart comparing F1 and BLEU-1 scores for different values of 'k'. The chart displays two sets of bars for each k value, one representing the F1 score (blue) and the other representing the BLEU-1 score (orange). The x-axis represents 'k values', and the y-axis represents the score. ### Components/Axes * **X-axis:** 'k values' with markers at 10, 20, 30, 40, and 50. * **Y-axis:** Numerical scale ranging from approximately 28 to 52, with no explicit label. * **Legend:** Located in the top-left corner, indicating that the blue bars represent 'F1' and the orange bars represent 'BLEU-1'. * **Gridlines:** Horizontal dashed gridlines are present. ### Detailed Analysis The chart presents the following data points: * **k = 10:** * F1 (blue): 30.29 * BLEU-1 (orange): 29.49 * **k = 20:** * F1 (blue): 39.11 * BLEU-1 (orange): 38.35 * **k = 30:** * F1 (blue): 43.86 * BLEU-1 (orange): 43.19 * **k = 40:** * F1 (blue): 50.03 * BLEU-1 (orange): 49.47 * **k = 50:** * F1 (blue): 47.76 * BLEU-1 (orange): 47.24 **Trend Verification:** * **F1 (blue):** Generally increases from k=10 to k=40, then decreases slightly at k=50. * **BLEU-1 (orange):** Generally increases from k=10 to k=40, then decreases slightly at k=50. ### Key Observations * Both F1 and BLEU-1 scores increase as 'k' increases from 10 to 40. * Both scores peak at k=40 and then slightly decrease at k=50. * The F1 score is consistently slightly higher than the BLEU-1 score for each 'k' value. ### Interpretation The chart suggests that increasing the value of 'k' initially improves both F1 and BLEU-1 scores, indicating better performance up to a point. However, beyond k=40, the performance starts to decline slightly, suggesting that there might be an optimal value for 'k' around 40. The consistent difference between F1 and BLEU-1 scores might indicate inherent differences in what these metrics capture, or it could be a characteristic of the specific model or task being evaluated. </details> (e) Adversarial Figure 3: Impact of memory retrieval parameter k across different task categories with GPT-4o-mini as the base model. While larger k values generally improve performance by providing richer historical context, the gains diminish beyond certain thresholds, suggesting a trade-off between context richness and effective information processing. This pattern is consistent across all evaluation categories, indicating the importance of balanced context retrieval for optimal performance. 4.5 Hyperparameter Analysis We conducted extensive experiments to analyze the impact of the memory retrieval parameter k, which controls the number of relevant memories retrieved for each interaction. As shown in Figure 3, we evaluated performance across different k values (10, 20, 30, 40, 50) on five categories of tasks using GPT-4o-mini as our base model. The results reveal an interesting pattern: while increasing k generally leads to improved performance, this improvement gradually plateaus and sometimes slightly decreases at higher values. This trend is particularly evident in Multi Hop and Open Domain tasks. The observation suggests a delicate balance in memory retrieval - while larger k values provide richer historical context for reasoning, they may also introduce noise and challenge the model’s capacity to process longer sequences effectively. Our analysis indicates that moderate k values strike an optimal balance between context richness and information processing efficiency. Table 4: Comparison of memory usage and retrieval time across different memory methods and scales. | Memory Size | Method | Memory Usage (MB) | Retrieval Time ( $\mu\text{s}$ ) | | --- | --- | --- | --- | | 1,000 | A-Mem | 1.46 | 0.31 0.30 | | MemoryBank [39] | 1.46 | 0.24 0.20 | | | ReadAgent [17] | 1.46 | 43.62 8.47 | | | 10,000 | A-Mem | 14.65 | 0.38 0.25 | | MemoryBank [39] | 14.65 | 0.26 0.13 | | | ReadAgent [17] | 14.65 | 484.45 93.86 | | | 100,000 | A-Mem | 146.48 | 1.40 0.49 | | MemoryBank [39] | 146.48 | 0.78 0.26 | | | ReadAgent [17] | 146.48 | 6,682.22 111.63 | | | 1,000,000 | A-Mem | 1464.84 | 3.70 0.74 | | MemoryBank [39] | 1464.84 | 1.91 0.31 | | | ReadAgent [17] | 1464.84 | 120,069.68 1,673.39 | | 4.6 Scaling Analysis To evaluate storage costs with accumulating memory, we examined the relationship between storage size and retrieval time across our A-Mem system and two baseline approaches: MemoryBank [39] and ReadAgent [17]. We evaluated these three memory systems with identical memory content across four scale points, increasing the number of entries by a factor of 10 at each step (from 1,000 to 10,000, 100,000, and finally 1,000,000 entries). The experimental results reveal key insights about our A-Mem system’s scaling properties: In terms of space complexity, all three systems exhibit identical linear memory usage scaling ( $O(N)$ ), as expected for vector-based retrieval systems. This confirms that A-Mem introduces no additional storage overhead compared to baseline approaches. For retrieval time, A-Mem demonstrates excellent efficiency with minimal increases as memory size grows. Even when scaling to 1 million memories, A-Mem ’s retrieval time increases only from 0.31 $\mu\text{s}$ to 3.70 $\mu\text{s}$ , representing exceptional performance. While MemoryBank shows slightly faster retrieval times, A-Mem maintains comparable performance while providing richer memory representations and functionality. Based on our space complexity and retrieval time analysis, we conclude that A-Mem ’s retrieval mechanisms maintain excellent efficiency even at large scales. The minimal growth in retrieval time across memory sizes addresses concerns about efficiency in large-scale memory systems, demonstrating that A-Mem provides a highly scalable solution for long-term conversation management. This unique combination of efficiency, scalability, and enhanced memory capabilities positions A-Mem as a significant advancement in building powerful and long-term memory mechanism for LLM Agents. 4.7 Memory Analysis We present the t-SNE visualization in Figure 4 of memory embeddings to demonstrate the structural advantages of our agentic memory system. Analyzing two dialogues sampled from long-term conversations in LoCoMo [22], we observe that A-Mem (shown in blue) consistently exhibits more coherent clustering patterns compared to the baseline system (shown in red). This structural organization is particularly evident in Dialogue 2, where well-defined clusters emerge in the central region, providing empirical evidence for the effectiveness of our memory evolution mechanism and contextual description generation. In contrast, the baseline memory embeddings display a more dispersed distribution, demonstrating that memories lack structural organization without our link generation and memory evolution components. These visualization results validate that A-Mem can autonomously maintain meaningful memory structures through dynamic evolution and linking mechanisms. More results can be seen in Appendix A.4. <details> <summary>x11.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot comparing two datasets, "A-mem" and "Base". The plot displays the distribution of data points for each dataset across a two-dimensional space, with no explicit x or y axis labels. The "A-mem" data points are represented in light blue, while the "Base" data points are in light red. ### Components/Axes * **X-axis:** No explicit label, but ranges approximately from -25 to 25. * **Y-axis:** No explicit label, but ranges approximately from -25 to 25. * **Legend:** Located in the top-left corner. * "A-mem": Light blue data points. * "Base": Light red data points. ### Detailed Analysis The scatter plot shows the distribution of "A-mem" and "Base" data points. * **A-mem (Light Blue):** The light blue points are more concentrated in the central region of the plot, with a higher density around the origin (0,0). They are also present, but less dense, in the outer regions. * **Base (Light Red):** The light red points are more evenly distributed across the entire plot area, including the outer regions. They appear to have a more uniform spread compared to the "A-mem" data. Approximate data point ranges: * **A-mem:** * X-axis: Mostly between -15 and 15. * Y-axis: Mostly between -15 and 15. * **Base:** * X-axis: Ranging from -25 to 25. * Y-axis: Ranging from -25 to 25. ### Key Observations * The "A-mem" data points are more clustered towards the center of the plot, indicating a higher concentration in that region. * The "Base" data points are more dispersed, suggesting a broader distribution across the plot area. * There is some overlap between the two datasets, particularly in the central region. ### Interpretation The scatter plot suggests that the "A-mem" dataset exhibits a stronger central tendency compared to the "Base" dataset. This could indicate that "A-mem" data points are more similar to each other or that they are influenced by a common factor that pulls them towards the center. The "Base" dataset, on the other hand, appears to be more diverse or influenced by a wider range of factors, resulting in a more dispersed distribution. The lack of axis labels makes it difficult to interpret the specific meaning of the x and y dimensions, but the relative distributions of the two datasets provide valuable insights into their underlying characteristics. </details> (a) Dialogue 1 <details> <summary>x12.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs Base ### Overview The image is a scatter plot showing the distribution of two datasets, labeled "A-mem" and "Base". The plot displays the data points in a two-dimensional space, with no explicit x or y axis labels. The data points are scattered across the plot, with some overlap between the two datasets. ### Components/Axes * **X-axis:** Ranges from approximately -20 to 20, with tick marks at -20, -10, 0, 10, and 20. No label is provided. * **Y-axis:** Ranges from approximately -20 to 30, with tick marks at -20, -10, 0, 10, 20, and 30. No label is provided. * **Legend (Top-Left):** * A-mem: Represented by light blue dots. * Base: Represented by light red dots. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points are scattered throughout the plot, with a higher concentration in the central region. * **Base (Light Red):** The light red data points are also scattered throughout the plot, with a distribution similar to the "A-mem" data, but perhaps slightly more dispersed. **Specific Data Point Analysis:** It is impossible to provide exact coordinates for each point due to the nature of a scatter plot and the lack of gridlines. However, we can describe the general distribution: * **A-mem:** * Most points are concentrated within the range of x = -10 to 10 and y = -10 to 20. * There are fewer points in the extreme corners of the plot. * **Base:** * Similar distribution to A-mem, but with slightly more points extending towards the edges of the plot. * The density of points appears slightly lower than A-mem in the central region. ### Key Observations * Both datasets ("A-mem" and "Base") exhibit a roughly similar distribution pattern. * There is significant overlap between the two datasets, suggesting that they are not easily separable in this two-dimensional space. * The central region of the plot appears to have a higher density of points for both datasets. ### Interpretation The scatter plot visualizes the relationship between two datasets, "A-mem" and "Base". The overlapping distributions suggest that the two datasets share similar characteristics or are influenced by similar factors. Without further context or axis labels, it is difficult to determine the specific meaning of the plot. However, the visualization provides a basis for comparing the two datasets and identifying potential patterns or clusters. The plot suggests that a simple linear separation of the two datasets is unlikely to be effective. </details> (b) Dialogue 2 Figure 4: T-SNE Visualization of Memory Embeddings Showing More Organized Distribution with A-Mem (blue) Compared to Base Memory (red) Across Different Dialogues. Base Memory represents A-Mem without link generation and memory evolution. 5 Conclusions In this work, we introduced A-Mem, a novel agentic memory system that enables LLM agents to dynamically organize and evolve their memories without relying on predefined structures. Drawing inspiration from the Zettelkasten method, our system creates an interconnected knowledge network through dynamic indexing and linking mechanisms that adapt to diverse real-world tasks. The system’s core architecture features autonomous generation of contextual descriptions for new memories and intelligent establishment of connections with existing memories based on shared attributes. Furthermore, our approach enables continuous evolution of historical memories by incorporating new experiences and developing higher-order attributes through ongoing interactions. Through extensive empirical evaluation across six foundation models, we demonstrated that A-Mem achieves superior performance compared to existing state-of-the-art baselines in long-term conversational tasks. Visualization analysis further validates the effectiveness of our memory organization approach. These results suggest that agentic memory systems can significantly enhance LLM agents’ ability to utilize long-term knowledge in complex environments. 6 Limitations While our agentic memory system achieves promising results, we acknowledge several areas for potential future exploration. First, although our system dynamically organizes memories, the quality of these organizations may still be influenced by the inherent capabilities of the underlying language models. Different LLMs might generate slightly different contextual descriptions or establish varying connections between memories. Additionally, while our current implementation focuses on text-based interactions, future work could explore extending the system to handle multimodal information, such as images or audio, which could provide richer contextual representations. References - [1] Sönke Ahrens. How to Take Smart Notes: One Simple Technique to Boost Writing, Learning and Thinking. Amazon, 2017. Second Edition. - [2] Anthropic. The claude 3 model family: Opus, sonnet, haiku. Anthropic, Mar 2024. Accessed May 2025. - [3] Anthropic. Claude 3.5 sonnet model card addendum. Technical report, Anthropic, 2025. Accessed May 2025. - [4] Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. Self-rag: Learning to retrieve, generate, and critique through self-reflection. arXiv preprint arXiv:2310.11511, 2023. - [5] Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005. - [6] Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens. In International conference on machine learning, pages 2206–2240. PMLR, 2022. - [7] Xiang Deng, Yu Gu, Boyuan Zheng, Shijie Chen, Sam Stevens, Boshi Wang, Huan Sun, and Yu Su. Mind2web: Towards a generalist agent for the web. Advances in Neural Information Processing Systems, 36:28091–28114, 2023. - [8] Khant Dev and Singh Taranjeet. mem0: The memory layer for ai agents. https://github.com/mem0ai/mem0, 2024. - [9] Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130, 2024. - [10] Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, and Haofen Wang. Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997, 2023. - [11] Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025. - [12] I. Ilin. Advanced rag techniques: An illustrated overview, 2023. - [13] Jihyoung Jang, Minseong Boo, and Hyounghun Kim. Conversation chronicles: Towards diverse temporal and relational dynamics in multi-session conversations. arXiv preprint arXiv:2310.13420, 2023. - [14] Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. Active retrieval augmented generation. arXiv preprint arXiv:2305.06983, 2023. - [15] David Kadavy. Digital Zettelkasten: Principles, Methods, & Examples. Google Books, May 2021. - [16] Jiho Kim, Woosog Chay, Hyeonji Hwang, Daeun Kyung, Hyunseung Chung, Eunbyeol Cho, Yohan Jo, and Edward Choi. Dialsim: A real-time simulator for evaluating long-term multi-party dialogue understanding of conversational agents. arXiv preprint arXiv:2406.13144, 2024. - [17] Kuang-Huei Lee, Xinyun Chen, Hiroki Furuta, John Canny, and Ian Fischer. A human-inspired reading agent with gist memory of very long contexts. arXiv preprint arXiv:2402.09727, 2024. - [18] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems, 33:9459–9474, 2020. - [19] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81, 2004. - [20] Xi Victoria Lin, Xilun Chen, Mingda Chen, Weijia Shi, Maria Lomeli, Rich James, Pedro Rodriguez, Jacob Kahn, Gergely Szilvasy, Mike Lewis, et al. Ra-dit: Retrieval-augmented dual instruction tuning. arXiv preprint arXiv:2310.01352, 2023. - [21] Zhiwei Liu, Weiran Yao, Jianguo Zhang, Liangwei Yang, Zuxin Liu, Juntao Tan, Prafulla K Choubey, Tian Lan, Jason Wu, Huan Wang, et al. Agentlite: A lightweight library for building and advancing task-oriented llm agent system. arXiv preprint arXiv:2402.15538, 2024. - [22] Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. Evaluating very long-term conversational memory of llm agents. arXiv preprint arXiv:2402.17753, 2024. - [23] Kai Mei, Zelong Li, Shuyuan Xu, Ruosong Ye, Yingqiang Ge, and Yongfeng Zhang. Aios: Llm agent operating system. arXiv e-prints, pp. arXiv–2403, 2024. - [24] Ali Modarressi, Ayyoob Imani, Mohsen Fayyaz, and Hinrich Schütze. Ret-llm: Towards a general read-write memory for large language models. arXiv preprint arXiv:2305.14322, 2023. - [25] Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G Patil, Ion Stoica, and Joseph E Gonzalez. Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560, 2023. - [26] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318, 2002. - [27] Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 11 2019. - [28] Aymeric Roucher, Albert Villanova del Moral, Thomas Wolf, Leandro von Werra, and Erik Kaunismäki. ‘smolagents‘: a smol library to build great agentic systems. https://github.com/huggingface/smolagents, 2025. - [29] Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, and Weizhu Chen. Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy. arXiv preprint arXiv:2305.15294, 2023. - [30] Zeru Shi, Kai Mei, Mingyu Jin, Yongye Su, Chaoji Zuo, Wenyue Hua, Wujiang Xu, Yujie Ren, Zirui Liu, Mengnan Du, et al. From commands to prompts: Llm-based semantic file system for aios. arXiv preprint arXiv:2410.11843, 2024. - [31] Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions. arXiv preprint arXiv:2212.10509, 2022. - [32] Bing Wang, Xinnian Liang, Jian Yang, Hui Huang, Shuangzhi Wu, Peihao Wu, Lu Lu, Zejun Ma, and Zhoujun Li. Enhancing large language model with self-controlled memory framework. arXiv preprint arXiv:2304.13343, 2023. - [33] Xingyao Wang, Boxuan Li, Yufan Song, Frank F Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, et al. Openhands: An open platform for ai software developers as generalist agents. arXiv preprint arXiv:2407.16741, 2024. - [34] Zhiruo Wang, Jun Araki, Zhengbao Jiang, Md Rizwan Parvez, and Graham Neubig. Learning to filter context for retrieval-augmented generation. arXiv preprint arXiv:2311.08377, 2023. - [35] Lilian Weng. Llm-powered autonomous agents. lilianweng.github.io, Jun 2023. - [36] J Xu. Beyond goldfish memory: Long-term open-domain conversation. arXiv preprint arXiv:2107.07567, 2021. - [37] Wenhao Yu, Hongming Zhang, Xiaoman Pan, Kaixin Ma, Hongwei Wang, and Dong Yu. Chain-of-note: Enhancing robustness in retrieval-augmented language models. arXiv preprint arXiv:2311.09210, 2023. - [38] Zichun Yu, Chenyan Xiong, Shi Yu, and Zhiyuan Liu. Augmentation-adapted retriever improves generalization of language models as generic plug-in. arXiv preprint arXiv:2305.17331, 2023. - [39] Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19724–19731, 2024. Contents 1. 1 Introduction 1. 2 Related Work 1. 2.1 Memory for LLM Agents 1. 2.2 Retrieval-Augmented Generation 1. 3 Methodolodgy 1. 3.1 Note Construction 1. 3.2 Link Generation 1. 3.3 Memory Evolution 1. 3.4 Retrieve Relative Memory 1. 4 Experiment 1. 4.1 Dataset and Evaluation 1. 4.2 Implementation Details 1. 4.3 Empricial Results 1. 4.4 Ablation Study 1. 4.5 Hyperparameter Analysis 1. 4.6 Scaling Analysis 1. 4.7 Memory Analysis 1. 5 Conclusions 1. 6 Limitations 1. A Experiment 1. A.1 Detailed Baselines Introduction 1. A.2 Evaluation Metric 1. A.3 Comparison Results 1. A.4 Memory Analysis 1. A.5 Hyperparameters setting 1. B Prompt Templates and Examples 1. B.1 Prompt Template of Note Construction 1. B.2 Prompt Template of Link Generation 1. B.3 Prompt Template of Memory Evolution 1. B.4 Examples of Q/A with A-Mem APPENDIX Appendix A Experiment A.1 Detailed Baselines Introduction LoCoMo [22] takes a direct approach by leveraging foundation models without memory mechanisms for question answering tasks. For each query, it incorporates the complete preceding conversation and questions into the prompt, evaluating the model’s reasoning capabilities. ReadAgent [17] tackles long-context document processing through a sophisticated three-step methodology: it begins with episode pagination to segment content into manageable chunks, followed by memory gisting to distill each page into concise memory representations, and concludes with interactive look-up to retrieve pertinent information as needed. MemoryBank [39] introduces an innovative memory management system that maintains and efficiently retrieves historical interactions. The system features a dynamic memory updating mechanism based on the Ebbinghaus Forgetting Curve theory, which intelligently adjusts memory strength according to time and significance. Additionally, it incorporates a user portrait building system that progressively refines its understanding of user personality through continuous interaction analysis. MemGPT [25] presents a novel virtual context management system drawing inspiration from traditional operating systems’ memory hierarchies. The architecture implements a dual-tier structure: a main context (analogous to RAM) that provides immediate access during LLM inference, and an external context (analogous to disk storage) that maintains information beyond the fixed context window. A.2 Evaluation Metric The F1 score represents the harmonic mean of precision and recall, offering a balanced metric that combines both measures into a single value. This metric is particularly valuable when we need to balance between complete and accurate responses: $$ \text{F1}=2\cdot\frac{\text{precision}\cdot\text{recall}}{\text{precision}+\text{recall}} \tag{11} $$ where $$ \text{precision}=\frac{\text{true positives}}{\text{true positives}+\text{false positives}} \tag{12} $$ $$ \text{recall}=\frac{\text{true positives}}{\text{true positives}+\text{false negatives}} \tag{13} $$ In question-answering systems, the F1 score serves a crucial role in evaluating exact matches between predicted and reference answers. This is especially important for span-based QA tasks, where systems must identify precise text segments while maintaining comprehensive coverage of the answer. BLEU-1 [26] provides a method for evaluating the precision of unigram matches between system outputs and reference texts: $$ \text{BLEU-1}=BP\cdot\exp(\tsum\slimits@_{n=1}^{1}w_{n}\log p_{n}) \tag{14} $$ where $$ BP=\begin{cases}1&\text{if }c>r\\ e^{1-r/c}&\text{if }c\leq r\end{cases} \tag{15} $$ $$ p_{n}=\frac{\tsum\slimits@_{i}\tsum\slimits@_{k}\min(h_{ik},m_{ik})}{\tsum\slimits@_{i}\tsum\slimits@_{k}h_{ik}} \tag{16} $$ Here, $c$ is candidate length, $r$ is reference length, $h_{ik}$ is the count of n-gram i in candidate k, and $m_{ik}$ is the maximum count in any reference. In QA, BLEU-1 evaluates the lexical precision of generated answers, particularly useful for generative QA systems where exact matching might be too strict. ROUGE-L [19] measures the longest common subsequence between the generated and reference texts. $$ \text{ROUGE-L}=\frac{(1+\beta^{2})R_{l}P_{l}}{R_{l}+\beta^{2}P_{l}} \tag{17} $$ $$ R_{l}=\frac{\text{LCS}(X,Y)}{|X|} \tag{18} $$ $$ P_{l}=\frac{\text{LCS}(X,Y)}{|Y|} \tag{19} $$ where $X$ is reference text, $Y$ is candidate text, and LCS is the Longest Common Subsequence. ROUGE-2 [19] calculates the overlap of bigrams between the generated and reference texts. $$ \text{ROUGE-2}=\frac{\tsum\slimits@_{\text{bigram}\in\text{ref}}\min(\text{Count}_{\text{ref}}(\text{bigram}),\text{Count}_{\text{cand}}(\text{bigram}))}{\tsum\slimits@_{\text{bigram}\in\text{ref}}\text{Count}_{\text{ref}}(\text{bigram})} \tag{20} $$ Both ROUGE-L and ROUGE-2 are particularly useful for evaluating the fluency and coherence of generated answers, with ROUGE-L focusing on sequence matching and ROUGE-2 on local word order. METEOR [5] computes a score based on aligned unigrams between the candidate and reference texts, considering synonyms and paraphrases. $$ \text{METEOR}=F_{\text{mean}}\cdot(1-\text{Penalty}) \tag{21} $$ $$ F_{\text{mean}}=\frac{10P\cdot R}{R+9P} \tag{22} $$ $$ \text{Penalty}=0.5\cdot(\frac{\text{ch}}{m})^{3} \tag{23} $$ where $P$ is precision, $R$ is recall, ch is number of chunks, and $m$ is number of matched unigrams. METEOR is valuable for QA evaluation as it considers semantic similarity beyond exact matching, making it suitable for evaluating paraphrased answers. SBERT Similarity [27] measures the semantic similarity between two texts using sentence embeddings. $$ \text{SBERT\_Similarity}=\cos(\text{SBERT}(x),\text{SBERT}(y)) \tag{24} $$ $$ \cos(a,b)=\frac{a\cdot b}{\|a\|\|b\|} \tag{25} $$ SBERT( $x$ ) represents the sentence embedding of text. SBERT Similarity is particularly useful for evaluating semantic understanding in QA systems, as it can capture meaning similarities even when the lexical overlap is low. Table 5: Experimental results on LoCoMo dataset of QA tasks across five categories (Multi Hop, Temporal, Open Domain, Single Hop, and Adversial) using different methods. Results are reported in ROUGE-2 and ROUGE-L scores, abbreviated to RGE-2 and RGE-L. The best performance is marked in bold, and our proposed method A-Mem (highlighted in gray) demonstrates competitive performance across six foundation language models. | Model | Method | Category | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | | | | | | | | | | RGE-2 | RGE-L | RGE-2 | RGE-L | RGE-2 | RGE-L | RGE-2 | RGE-L | RGE-2 | RGE-L | | | | | GPT | 4o-mini | LoCoMo | 9.64 | 23.92 | 2.01 | 18.09 | 3.40 | 11.58 | 26.48 | 40.20 | 60.46 | 69.59 | | ReadAgent | 2.47 | 9.45 | 0.95 | 13.12 | 0.55 | 5.76 | 2.99 | 9.92 | 6.66 | 9.79 | | | | MemoryBank | 1.18 | 5.43 | 0.52 | 9.64 | 0.97 | 5.77 | 1.64 | 6.63 | 4.55 | 7.35 | | | | MemGPT | 10.58 | 25.60 | 4.76 | 25.22 | 0.76 | 9.14 | 28.44 | 42.24 | 36.62 | 43.75 | | | | A-Mem | 10.61 | 25.86 | 21.39 | 44.27 | 3.42 | 12.09 | 29.50 | 45.18 | 42.62 | 50.04 | | | | 4o | LoCoMo | 11.53 | 30.65 | 1.68 | 8.17 | 3.21 | 16.33 | 45.42 | 63.86 | 45.13 | 52.67 | | | ReadAgent | 3.91 | 14.36 | 0.43 | 3.96 | 0.52 | 8.58 | 4.75 | 13.41 | 4.24 | 6.81 | | | | MemoryBank | 1.84 | 7.36 | 0.36 | 2.29 | 2.13 | 6.85 | 3.02 | 9.35 | 1.22 | 4.41 | | | | MemGPT | 11.55 | 30.18 | 4.66 | 15.83 | 3.27 | 14.02 | 43.27 | 62.75 | 28.72 | 35.08 | | | | A-Mem | 12.76 | 31.71 | 9.82 | 25.04 | 6.09 | 16.63 | 33.67 | 50.31 | 30.31 | 36.34 | | | | Qwen2.5 | 1.5b | LoCoMo | 1.39 | 9.24 | 0.00 | 4.68 | 3.42 | 10.59 | 3.25 | 11.15 | 35.10 | 43.61 | | ReadAgent | 0.74 | 7.14 | 0.10 | 2.81 | 3.05 | 12.63 | 1.47 | 7.88 | 20.73 | 27.82 | | | | MemoryBank | 1.51 | 11.18 | 0.14 | 5.39 | 1.80 | 8.44 | 5.07 | 13.72 | 29.24 | 36.95 | | | | MemGPT | 1.16 | 11.35 | 0.00 | 7.88 | 2.87 | 14.62 | 2.18 | 9.82 | 23.96 | 31.69 | | | | A-Mem | 4.88 | 17.94 | 5.88 | 27.23 | 3.44 | 16.87 | 12.32 | 24.38 | 36.32 | 46.60 | | | | 3b | LoCoMo | 0.49 | 4.83 | 0.14 | 3.20 | 1.31 | 5.38 | 1.97 | 6.98 | 12.66 | 17.10 | | | ReadAgent | 0.08 | 4.08 | 0.00 | 1.96 | 1.26 | 6.19 | 0.73 | 4.34 | 7.35 | 10.64 | | | | MemoryBank | 0.43 | 3.76 | 0.05 | 1.61 | 0.24 | 6.32 | 1.03 | 4.22 | 9.55 | 13.41 | | | | MemGPT | 0.69 | 5.55 | 0.05 | 3.17 | 1.90 | 7.90 | 2.05 | 7.32 | 10.46 | 14.39 | | | | A-Mem | 2.91 | 12.42 | 8.11 | 27.74 | 1.51 | 7.51 | 8.80 | 17.57 | 21.39 | 27.98 | | | | Llama 3.2 | 1b | LoCoMo | 2.51 | 11.48 | 0.44 | 8.25 | 1.69 | 13.06 | 2.94 | 13.00 | 39.85 | 52.74 | | ReadAgent | 0.53 | 6.49 | 0.00 | 4.62 | 5.47 | 14.29 | 1.19 | 8.03 | 34.52 | 45.55 | | | | MemoryBank | 2.96 | 13.57 | 0.23 | 10.53 | 4.01 | 18.38 | 6.41 | 17.66 | 41.15 | 53.31 | | | | MemGPT | 1.82 | 9.91 | 0.06 | 6.56 | 2.13 | 11.36 | 2.00 | 10.37 | 38.59 | 50.31 | | | | A-Mem | 4.82 | 19.31 | 1.84 | 20.47 | 5.99 | 18.49 | 14.82 | 29.78 | 46.76 | 60.23 | | | | 3b | LoCoMo | 0.98 | 7.22 | 0.03 | 4.45 | 2.36 | 11.39 | 2.85 | 8.45 | 25.47 | 30.26 | | | ReadAgent | 2.47 | 1.78 | 3.01 | 3.01 | 5.07 | 5.22 | 3.25 | 2.51 | 15.78 | 14.01 | | | | MemoryBank | 1.83 | 6.96 | 0.25 | 3.41 | 0.43 | 4.43 | 2.73 | 7.83 | 14.64 | 18.59 | | | | MemGPT | 0.72 | 5.39 | 0.11 | 2.85 | 0.61 | 5.74 | 1.45 | 4.42 | 16.62 | 21.47 | | | | A-Mem | 6.02 | 17.62 | 7.93 | 27.97 | 5.38 | 13.00 | 16.89 | 28.55 | 35.48 | 42.25 | | | Table 6: Experimental results on LoCoMo dataset of QA tasks across five categories (Multi Hop, Temporal, Open Domain, Single Hop, and Adversial) using different methods. Results are reported in METEOR and SBERT Similarity scores, abbreviated to ME and SBERT. The best performance is marked in bold, and our proposed method A-Mem (highlighted in gray) demonstrates competitive performance across six foundation language models. | Model | Method | Category | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | | | | | | | | | | ME | SBERT | ME | SBERT | ME | SBERT | ME | SBERT | ME | SBERT | | | | | GPT | 4o-mini | LoCoMo | 15.81 | 47.97 | 7.61 | 52.30 | 8.16 | 35.00 | 40.42 | 57.78 | 63.28 | 71.93 | | ReadAgent | 5.46 | 28.67 | 4.76 | 45.07 | 3.69 | 26.72 | 8.01 | 26.78 | 8.38 | 15.20 | | | | MemoryBank | 3.42 | 21.71 | 4.07 | 37.58 | 4.21 | 23.71 | 5.81 | 20.76 | 6.24 | 13.00 | | | | MemGPT | 15.79 | 49.33 | 13.25 | 61.53 | 4.59 | 32.77 | 41.40 | 58.19 | 39.16 | 47.24 | | | | A-Mem | 16.36 | 49.46 | 23.43 | 70.49 | 8.36 | 38.48 | 42.32 | 59.38 | 45.64 | 53.26 | | | | 4o | LoCoMo | 16.34 | 53.82 | 7.21 | 32.15 | 8.98 | 43.72 | 53.39 | 73.40 | 47.72 | 56.09 | | | ReadAgent | 7.86 | 37.41 | 3.76 | 26.22 | 4.42 | 30.75 | 9.36 | 31.37 | 5.47 | 12.34 | | | | MemoryBank | 3.22 | 26.23 | 2.29 | 23.49 | 4.18 | 24.89 | 6.64 | 23.90 | 2.93 | 10.01 | | | | MemGPT | 16.64 | 55.12 | 12.68 | 35.93 | 7.78 | 37.91 | 52.14 | 72.83 | 31.15 | 39.08 | | | | A-Mem | 17.53 | 55.96 | 13.10 | 45.40 | 10.62 | 38.87 | 41.93 | 62.47 | 32.34 | 40.11 | | | | Qwen2.5 | 1.5b | LoCoMo | 4.99 | 32.23 | 2.86 | 34.03 | 5.89 | 35.61 | 8.57 | 29.47 | 40.53 | 50.49 | | ReadAgent | 3.67 | 28.20 | 1.88 | 27.27 | 8.97 | 35.13 | 5.52 | 26.33 | 24.04 | 34.12 | | | | MemoryBank | 5.57 | 35.40 | 2.80 | 32.47 | 4.27 | 33.85 | 10.59 | 32.16 | 32.93 | 42.83 | | | | MemGPT | 5.40 | 35.64 | 2.35 | 39.04 | 7.68 | 40.36 | 7.07 | 30.16 | 27.24 | 40.63 | | | | A-Mem | 9.49 | 43.49 | 11.92 | 61.65 | 9.11 | 42.58 | 19.69 | 41.93 | 40.64 | 52.44 | | | | 3b | LoCoMo | 2.00 | 24.37 | 1.92 | 25.24 | 3.45 | 25.38 | 6.00 | 21.28 | 16.67 | 23.14 | | | ReadAgent | 1.78 | 21.10 | 1.69 | 20.78 | 4.43 | 25.15 | 3.37 | 18.20 | 10.46 | 17.39 | | | | MemoryBank | 2.37 | 17.81 | 2.22 | 21.93 | 3.86 | 20.65 | 3.99 | 16.26 | 15.49 | 20.77 | | | | MemGPT | 3.74 | 24.31 | 2.25 | 27.67 | 6.44 | 29.59 | 6.24 | 22.40 | 13.19 | 20.83 | | | | A-Mem | 6.25 | 33.72 | 14.04 | 62.54 | 6.56 | 30.60 | 15.98 | 33.98 | 27.36 | 33.72 | | | | Llama 3.2 | 1b | LoCoMo | 5.77 | 38.02 | 3.38 | 45.44 | 6.20 | 42.69 | 9.33 | 34.19 | 46.79 | 60.74 | | ReadAgent | 2.97 | 29.26 | 1.31 | 26.45 | 7.13 | 39.19 | 5.36 | 26.44 | 42.39 | 54.35 | | | | MemoryBank | 6.77 | 39.33 | 4.43 | 45.63 | 7.76 | 42.81 | 13.01 | 37.32 | 50.43 | 60.81 | | | | MemGPT | 5.10 | 32.99 | 2.54 | 41.81 | 3.26 | 35.99 | 6.62 | 30.68 | 45.00 | 61.33 | | | | A-Mem | 9.01 | 45.16 | 7.50 | 54.79 | 8.30 | 43.42 | 22.46 | 47.07 | 53.72 | 68.00 | | | | 3b | LoCoMo | 3.69 | 27.94 | 2.96 | 20.40 | 6.46 | 32.17 | 6.58 | 22.92 | 29.02 | 35.74 | | | ReadAgent | 1.21 | 17.40 | 2.33 | 12.02 | 3.39 | 19.63 | 2.46 | 14.63 | 14.37 | 21.25 | | | | MemoryBank | 3.84 | 25.06 | 2.73 | 13.65 | 3.05 | 21.08 | 6.35 | 22.02 | 17.14 | 24.39 | | | | MemGPT | 2.78 | 22.06 | 2.21 | 14.97 | 3.63 | 23.18 | 3.47 | 17.81 | 20.50 | 26.87 | | | | A-Mem | 9.74 | 39.32 | 13.19 | 59.70 | 8.09 | 32.27 | 24.30 | 42.86 | 39.74 | 46.76 | | | Table 7: Experimental results on LoCoMo dataset of QA tasks across five categories (Multi Hop, Temporal, Open Domain, Single Hop, and Adversial) using different methods. Results are reported in F1 and BLEU-1 (%) scores with different foundation models. | Method | Category | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | | | | | | | | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | F1 | BLEU-1 | | | DeepSeek-R1-32B | | | | | | | | | | | | LoCoMo | 8.58 | 6.48 | 4.79 | 4.35 | 12.96 | 12.52 | 10.72 | 8.20 | 21.40 | 20.23 | | MemGPT | 8.28 | 6.25 | 5.45 | 4.97 | 10.97 | 9.09 | 11.34 | 9.03 | 30.77 | 29.23 | | A-Mem | 15.02 | 10.64 | 14.64 | 11.01 | 14.81 | 12.82 | 15.37 | 12.30 | 27.92 | 27.19 | | Claude 3.0 Haiku | | | | | | | | | | | | LoCoMo | 4.56 | 3.33 | 0.82 | 0.59 | 2.86 | 3.22 | 3.56 | 3.24 | 3.46 | 3.42 | | MemGPT | 7.65 | 6.36 | 1.65 | 1.26 | 7.41 | 6.64 | 8.60 | 7.29 | 7.66 | 7.37 | | A-Mem | 19.28 | 14.69 | 16.65 | 12.23 | 11.85 | 9.61 | 34.72 | 30.05 | 35.99 | 34.87 | | Claude 3.5 Haiku | | | | | | | | | | | | LoCoMo | 11.34 | 8.21 | 3.29 | 2.69 | 3.79 | 3.58 | 14.01 | 12.57 | 7.37 | 7.12 | | MemGPT | 8.27 | 6.55 | 3.99 | 2.76 | 4.71 | 4.48 | 16.52 | 14.89 | 5.64 | 5.45 | | A-Mem | 29.70 | 23.19 | 31.54 | 27.53 | 11.42 | 9.47 | 42.60 | 37.41 | 13.65 | 12.71 | A.3 Comparison Results Our comprehensive evaluation using ROUGE-2, ROUGE-L, METEOR, and SBERT metrics demonstrates that A-Mem achieves superior performance while maintaining remarkable computational efficiency. Through extensive empirical testing across various model sizes and task categories, we have established A-Mem as a more effective approach compared to existing baselines, supported by several compelling findings. In our analysis of non-GPT models, specifically Qwen2.5 and Llama 3.2, A-Mem consistently outperforms all baseline approaches across all metrics. The Multi-Hop category showcases particularly striking results, where Qwen2.5-15b with A-Mem achieves a ROUGE-L score of 27.23, dramatically surpassing LoComo’s 4.68 and ReadAgent’s 2.81 - representing a nearly six-fold improvement. This pattern of superiority extends consistently across METEOR and SBERT scores. When examining GPT-based models, our results reveal an interesting pattern. While LoComo and MemGPT demonstrate strong capabilities in Open Domain and Adversarial tasks, A-Mem shows remarkable superiority in Multi-Hop reasoning tasks. Using GPT-4o-mini, A-Mem achieves a ROUGE-L score of 44.27 in Multi-Hop tasks, more than doubling LoComo’s 18.09. This significant advantage maintains consistency across other metrics, with METEOR scores of 23.43 versus 7.61 and SBERT scores of 70.49 versus 52.30. The significance of these results is amplified by A-Mem ’s exceptional computational efficiency. Our approach requires only 1,200-2,500 tokens, compared to the substantial 16,900 tokens needed by LoComo and MemGPT. This efficiency stems from two key architectural innovations: First, our novel agentic memory architecture creates interconnected memory networks through atomic notes with rich contextual descriptions, enabling more effective capture and utilization of information relationships. Second, our selective top-k retrieval mechanism facilitates dynamic memory evolution and structured organization. The effectiveness of these innovations is particularly evident in complex reasoning tasks, as demonstrated by the consistently strong Multi-Hop performance across all evaluation metrics. Besides, we also show the experimental results with different foundational models including DeepSeek-R1-32B [11], Claude 3.0 Haiku [2] and Claude 3.5 Haiku [3]. A.4 Memory Analysis In addition to the memory visualizations of the first two dialogues shown in the main text, we present additional visualizations in Fig. 5 that demonstrate the structural advantages of our agentic memory system. Through analysis of two dialogues sampled from long-term conversations in LoCoMo [22], we observe that A-Mem (shown in blue) consistently produces more coherent clustering patterns compared to the baseline system (shown in red). This structural organization is particularly evident in Dialogue 2, where distinct clusters emerge in the central region, providing empirical support for the effectiveness of our memory evolution mechanism and contextual description generation. In contrast, the baseline memory embeddings exhibit a more scattered distribution, indicating that memories lack structural organization without our link generation and memory evolution components. These visualizations validate that A-Mem can autonomously maintain meaningful memory structures through its dynamic evolution and linking mechanisms. <details> <summary>x13.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot comparing two datasets, "A-mem" and "Base". The plot displays the distribution of data points for each dataset across a two-dimensional space. The x and y axes range from approximately -30 to 40. ### Components/Axes * **X-axis:** Ranges from -20 to 20, with no explicit label. * **Y-axis:** Ranges from -30 to 40, with no explicit label. * **Legend (Top-Right):** * "A-mem": Represented by light blue dots. * "Base": Represented by light red dots. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points are scattered across the plot. There appear to be clusters of higher density in certain regions, particularly around x=0, y=20 and x=10, y=-10. * **Base (Light Red):** The light red data points are also scattered across the plot. The distribution appears relatively uniform, with no immediately obvious clusters. ### Key Observations * Both datasets cover a similar spatial area. * There is significant overlap between the two datasets. * The "A-mem" dataset appears to have some areas of higher density clustering compared to the "Base" dataset. ### Interpretation The scatter plot visualizes the distribution of two datasets, "A-mem" and "Base," in a two-dimensional space. The overlap between the datasets suggests that they share some similarities or common characteristics. The clustering observed in the "A-mem" dataset may indicate specific patterns or relationships within that dataset that are not present in the "Base" dataset. Without further context or information about the nature of the data, it is difficult to draw more specific conclusions. </details> (a) Dialogue 3 <details> <summary>x14.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs Base ### Overview The image is a scatter plot comparing two datasets, "A-mem" and "Base". The plot displays the distribution of data points for each dataset across a two-dimensional space. The x and y axes are not explicitly labeled, but the data points are clustered in distinct regions. ### Components/Axes * **X-axis:** Ranges from approximately -40 to 30. * **Y-axis:** Ranges from approximately -30 to 30. * **Legend (Top-Right):** * A-mem (Light Blue) * Base (Light Red) ### Detailed Analysis The scatter plot shows the distribution of two datasets, A-mem (light blue) and Base (light red). The data points are clustered, suggesting some underlying structure or grouping within the data. * **A-mem (Light Blue):** The light blue data points are clustered in several distinct regions. These clusters appear to be more concentrated than the Base data points. The clusters are located approximately at: * (-20, -5) * (-5, -15) * (0, 25) * (5, 0) * (15, -15) * (20, 0) * **Base (Light Red):** The light red data points are more dispersed compared to the A-mem data. They form a broader distribution, with a higher density around the clusters of A-mem. ### Key Observations * The A-mem data exhibits clear clustering, indicating potential groupings or categories within that dataset. * The Base data is more dispersed, suggesting a more uniform distribution or less distinct groupings. * The clusters of A-mem appear to be surrounded by a higher density of Base data points. ### Interpretation The scatter plot suggests that the "A-mem" dataset has a more structured organization compared to the "Base" dataset. The clustering of "A-mem" indicates that its data points may belong to distinct categories or groups. The "Base" dataset, on the other hand, appears to be more uniformly distributed, possibly representing a broader or less structured set of data. The higher density of "Base" data points around the "A-mem" clusters could indicate a relationship or interaction between the two datasets. The plot could be visualizing the results of a dimensionality reduction technique, where the original high-dimensional data is projected onto a two-dimensional space for visualization. </details> (b) Dialogue 4 <details> <summary>x15.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot comparing two datasets, labeled "A-mem" and "Base". The plot displays the distribution of data points for each dataset across a two-dimensional space. The x and y axes are not explicitly labeled, but they range from approximately -40 to 40 and -30 to 30, respectively. ### Components/Axes * **X-axis:** Ranges from approximately -40 to 40, with tick marks at -20, 0, 20, and 40. * **Y-axis:** Ranges from approximately -30 to 30, with tick marks at -20, -10, 0, 10, 20, and 30. * **Legend (top-left):** * "A-mem": Represented by light blue data points. * "Base": Represented by light red data points. ### Detailed Analysis * **A-mem (light blue):** The light blue data points are scattered across the plot, with a higher concentration in the central region. * **Base (light red):** The light red data points are also scattered across the plot, with a distribution that appears similar to "A-mem" but possibly more dispersed. ### Key Observations * Both datasets exhibit a roughly circular distribution centered around the origin (0,0). * The "Base" dataset appears to have a slightly wider spread than the "A-mem" dataset. * There is significant overlap between the two datasets. ### Interpretation The scatter plot visualizes the distribution of two datasets, "A-mem" and "Base," in a two-dimensional space. The overlapping distributions suggest that the two datasets share some similarities, but the slightly wider spread of the "Base" dataset may indicate some differences. Without further context or axis labels, it is difficult to determine the specific meaning of these distributions. However, the plot provides a visual comparison of the two datasets and highlights their relative similarities and differences in terms of spatial distribution. </details> (c) Dialogue 5 <details> <summary>x16.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot displaying two data sets, labeled "A-mem" and "Base". The data points are distributed in a roughly circular pattern, centered around the origin (0,0). The plot shows the spatial distribution of these two datasets, with some overlap. ### Components/Axes * **X-axis:** Ranges from approximately -25 to 25. No explicit label is provided. * **Y-axis:** Ranges from approximately -35 to 35. No explicit label is provided. * **Legend:** Located in the top-right corner. * "A-mem": Represented by light blue data points. * "Base": Represented by light red data points. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points are scattered throughout the plot, with a slightly higher concentration in the upper half of the circular distribution. * **Base (Light Red):** The light red data points are also scattered throughout the plot, appearing to fill in the spaces between the light blue points. * **Distribution:** Both datasets appear to be distributed in a roughly circular shape, centered around the origin. There is significant overlap between the two datasets. * **Specific Data Points:** * The highest Y-value for A-mem is approximately 30. * The lowest Y-value for A-mem is approximately -30. * The highest X-value for A-mem is approximately 25. * The lowest X-value for A-mem is approximately -25. * The highest Y-value for Base is approximately 35. * The lowest Y-value for Base is approximately -35. * The highest X-value for Base is approximately 25. * The lowest X-value for Base is approximately -25. ### Key Observations * The two datasets, "A-mem" and "Base," are intermixed, suggesting a degree of similarity or correlation in their underlying characteristics. * The circular distribution indicates that the data points are clustered around a central point, with no clear directional bias. * The absence of axis labels makes it difficult to interpret the specific meaning of the X and Y axes. ### Interpretation The scatter plot visualizes the relationship between two datasets, "A-mem" and "Base." The overlapping distribution suggests that the two datasets share some common characteristics or underlying factors. Without axis labels, it's impossible to determine the specific variables being compared. The plot could represent the results of a dimensionality reduction technique, such as t-SNE or PCA, where the original high-dimensional data has been projected onto two dimensions for visualization. The clustering around the origin might indicate that the data points are relatively similar to each other, with no strong outliers or distinct subgroups. </details> (d) Dialogue 6 <details> <summary>x17.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs Base ### Overview The image is a scatter plot comparing two datasets, labeled "A-mem" and "Base". The plot displays the distribution of data points for each dataset across a two-dimensional space. The x and y axes range from approximately -25 to 25 and -40 to 40, respectively. ### Components/Axes * **X-axis:** Ranges from -20 to 20, with tick marks at -20 and 0 and 20. * **Y-axis:** Ranges from -40 to 40, with tick marks at -40, -30, -20, -10, 0, 10, 20, 30, and 40. * **Legend:** Located in the top-right corner. * "A-mem" is represented by light blue dots. * "Base" is represented by light red dots. ### Detailed Analysis * **A-mem (Light Blue):** The light blue dots are scattered throughout the plot, with higher concentrations in certain areas, such as around (-5, -30), (10, -5), and (-5, 20). * **Base (Light Red):** The light red dots are also scattered throughout the plot, generally appearing more uniformly distributed compared to the "A-mem" data. * **Distribution:** Both datasets appear to be centered around the origin (0, 0), with data points spreading out in all directions. ### Key Observations * The "A-mem" data appears to have some clustering, while the "Base" data is more evenly distributed. * There is significant overlap between the two datasets, indicating some degree of similarity in their distributions. * The range of both datasets is approximately the same, spanning from -25 to 25 on the x-axis and -40 to 40 on the y-axis. ### Interpretation The scatter plot visualizes the distribution of two datasets, "A-mem" and "Base," in a two-dimensional space. The clustering observed in the "A-mem" data suggests that there may be underlying patterns or groupings within that dataset. The more uniform distribution of the "Base" data indicates a lack of such distinct patterns. The overlap between the two datasets suggests that they share some common characteristics, but the differences in their distributions imply that they also have unique features. The plot could be used to compare the performance or characteristics of two different models or algorithms, with "A-mem" potentially representing a model with memory or attention mechanisms and "Base" representing a baseline model. </details> (e) Dialogue 7 <details> <summary>x18.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot displaying the distribution of two datasets, labeled "A-mem" and "Base". The plot shows the data points scattered across a two-dimensional space, with the x and y axes ranging from approximately -30 to 30. The "A-mem" data points are colored light blue, while the "Base" data points are colored light red. ### Components/Axes * **X-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Y-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Legend:** Located in the top-right corner. * "A-mem": Represented by light blue data points. * "Base": Represented by light red data points. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points appear to be clustered in certain regions of the plot, forming denser groups. * **Base (Light Red):** The light red data points are more evenly distributed across the plot, with a slightly higher density in the central region. ### Key Observations * The "A-mem" data points show a tendency to cluster, suggesting some underlying structure or relationship within that dataset. * The "Base" data points are more dispersed, indicating a more uniform distribution across the space. * There is some overlap between the two datasets, but the clustering of "A-mem" allows for visual differentiation. ### Interpretation The scatter plot visualizes the distribution of two datasets, "A-mem" and "Base," in a two-dimensional space. The clustering observed in the "A-mem" dataset suggests that its data points share some common characteristics or relationships that are not as apparent in the "Base" dataset. The more uniform distribution of "Base" data points could indicate a lack of strong correlations or a more random distribution. The plot allows for a visual comparison of the two datasets and highlights the differences in their spatial distribution. </details> (f) Dialogue 8 <details> <summary>x19.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs. Base ### Overview The image is a scatter plot comparing two datasets, labeled "A-mem" and "Base." The plot displays the distribution of data points for each dataset across a two-dimensional space. The x and y axes range from approximately -30 to 30. ### Components/Axes * **X-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Y-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Legend:** Located in the top-left corner. * "A-mem": Represented by light blue data points. * "Base": Represented by light red data points. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points appear to be clustered in several regions. There are clusters around (-10, 15), (10, -10), (-10, -10), and (10, 15). * **Base (Light Red):** The light red data points are more evenly distributed across the space, with a higher density towards the edges of the plot. ### Key Observations * The "A-mem" dataset shows distinct clustering, suggesting some underlying structure or grouping within the data. * The "Base" dataset appears more uniformly distributed, indicating a lack of strong clustering. * There is some overlap between the two datasets, particularly in the central region of the plot. ### Interpretation The scatter plot visualizes the distribution of two datasets, "A-mem" and "Base," in a two-dimensional space. The clustering observed in the "A-mem" dataset suggests that its data points share some common characteristics or relationships, while the more uniform distribution of the "Base" dataset indicates a lack of such structure. The plot allows for a visual comparison of the two datasets and highlights their differences in distribution patterns. The clustering of "A-mem" could indicate that this data is more structured or has more dependencies than the "Base" data. </details> (g) Dialogue 9 <details> <summary>x20.png Details</summary>  ### Visual Description ## Scatter Plot: A-mem vs Base ### Overview The image is a scatter plot comparing two datasets, labeled "A-mem" (shown in light blue) and "Base" (shown in light red). The plot displays the distribution of data points for each dataset across a two-dimensional space, with x and y axes ranging from approximately -30 to 30. The data points appear to be scattered somewhat randomly, with some areas showing a higher density of points. ### Components/Axes * **X-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Y-axis:** Ranges from -30 to 30, with tick marks at -30, -20, -10, 0, 10, 20, and 30. * **Legend:** Located in the top-left corner. * "A-mem": Represented by light blue data points. * "Base": Represented by light red data points. ### Detailed Analysis * **A-mem (Light Blue):** The light blue data points are scattered throughout the plot. The density of points appears to be relatively uniform across the space, with a slight concentration around the center. * **Base (Light Red):** The light red data points are also scattered throughout the plot. Similar to "A-mem", the density of points seems relatively uniform, with a slight concentration around the center. ### Key Observations * Both datasets, "A-mem" and "Base", exhibit a similar distribution pattern. * There is no clear clustering or separation between the two datasets. * The data points are concentrated more towards the center of the plot. ### Interpretation The scatter plot suggests that the "A-mem" and "Base" datasets have similar characteristics in the two-dimensional space represented by the x and y axes. The lack of distinct clusters or separation indicates that the two datasets are not easily distinguishable based on these two dimensions. The concentration of points towards the center might indicate a central tendency or a higher probability of data points occurring in that region for both datasets. Further analysis with additional dimensions or different visualization techniques might be needed to identify any significant differences between the two datasets. </details> (h) Dialogue 10 Figure 5: T-SNE Visualization of Memory Embeddings Showing More Organized Distribution with A-Mem (blue) Compared to Base Memory (red) Across Different Dialogues. Base Memory represents A-Mem without link generation and memory evolution. A.5 Hyperparameters setting All hyperparameter k values are presented in Table 8. For models that have already achieved state-of-the-art (SOTA) performance with k=10, we maintain this value without further tuning. Table 8: Selection of k values in retriever across specific categories and model choices. | Model | Multi Hop | Temporal | Open Domain | Single Hop | Adversial | | --- | --- | --- | --- | --- | --- | | GPT-4o-mini | 40 | 40 | 50 | 50 | 40 | | GPT-4o | 40 | 40 | 50 | 50 | 40 | | Qwen2.5-1.5b | 10 | 10 | 10 | 10 | 10 | | Qwen2.5-3b | 10 | 10 | 50 | 10 | 10 | | Llama3.2-1b | 10 | 10 | 10 | 10 | 10 | | Llama3.2-3b | 10 | 20 | 10 | 10 | 10 | Appendix B Prompt Templates and Examples B.1 Prompt Template of Note Construction The prompt template in Note Construction: $P_{s1}$ Generate a structured analysis of the following content by: 1. Identifying the most salient keywords (focus on nouns, verbs, and key concepts) 2. Extracting core themes and contextual elements 3. Creating relevant categorical tags Format the response as a JSON object: { "keywords": [ // several specific, distinct keywords that capture key concepts and terminology // Order from most to least important // Don’t include keywords that are the name of the speaker or time // At least three keywords, but don’t be too redundant. ], "context": // one sentence summarizing: // - Main topic/domain // - Key arguments/points // - Intended audience/purpose , "tags": [ // several broad categories/themes for classification // Include domain, format, and type tags // At least three tags, but don’t be too redundant. ] } Content for analysis: B.2 Prompt Template of Link Generation The prompt template in Link Generation: $P_{s2}$ You are an AI memory evolution agent responsible for managing and evolving a knowledge base. Analyze the the new memory note according to keywords and context, also with their several nearest neighbors memory. The new memory context: {context} content: {content} keywords: {keywords} The nearest neighbors memories: {nearest_neighbors_memories} Based on this information, determine: Should this memory be evolved? Consider its relationships with other memories. B.3 Prompt Template of Memory Evolution The prompt template in Memory Evolution: $P_{s3}$ You are an AI memory evolution agent responsible for managing and evolving a knowledge base. Analyze the the new memory note according to keywords and context, also with their several nearest neighbors memory. Make decisions about its evolution. The new memory context:{context} content: {content} keywords: {keywords} The nearest neighbors memories:{nearest_neighbors_memories} Based on this information, determine: 1. What specific actions should be taken (strengthen, update_neighbor)? 1.1 If choose to strengthen the connection, which memory should it be connected to? Can you give the updated tags of this memory? 1.2 If choose to update neighbor, you can update the context and tags of these memories based on the understanding of these memories. Tags should be determined by the content of these characteristic of these memories, which can be used to retrieve them later and categorize them. All the above information should be returned in a list format according to the sequence: [[new_memory],[neighbor_memory_1], ...[neighbor_memory_n]] These actions can be combined. Return your decision in JSON format with the following structure: {{ "should_evolve": true/false, "actions": ["strengthen", "merge", "prune"], "suggested_connections": ["neighbor_memory_ids"], "tags_to_update": ["tag_1",..."tag_n"], "new_context_neighborhood": ["new context",...,"new context"], "new_tags_neighborhood": [["tag_1",...,"tag_n"],...["tag_1",...,"tag_n"]], }} B.4 Examples of Q/A with A-Mem Example: Question 686: Which hobby did Dave pick up in October 2023? Prediction: photography Reference: photography talk start time:10:54 am on 17 November, 2023 memory content: Speaker Davesays : Hey Calvin, long time no talk! A lot has happened. I’ve taken up photography and it’s been great - been taking pics of the scenery around here which is really cool. memory context: The main topic is the speaker’s new hobby of photography, highlighting their enjoyment of capturing local scenery, aimed at engaging a friend in conversation about personal experiences. memory keywords: [’photography’, ’scenery’, ’conversation’, ’experience’, ’hobby’] memory tags: [’hobby’, ’photography’, ’personal development’, ’conversation’, ’leisure’] talk start time:6:38 pm on 21 July, 2023 memory content: Speaker Calvinsays : Thanks, Dave! It feels great having my own space to work in. I’ve been experimenting with different genres lately, pushing myself out of my comfort zone. Adding electronic elements to my songs gives them a fresh vibe. It’s been an exciting process of self-discovery and growth! memory context: The speaker discusses their creative process in music, highlighting experimentation with genres and the incorporation of electronic elements for personal growth and artistic evolution. memory keywords: [’space’, ’experimentation’, ’genres’, ’electronic’, ’self-discovery’, ’growth’] memory tags: [’music’, ’creativity’, ’self-improvement’, ’artistic expression’] NeurIPS Paper Checklist The checklist is designed to encourage best practices for responsible machine learning research, addressing issues of reproducibility, transparency, research ethics, and societal impact. Do not remove the checklist: The papers not including the checklist will be desk rejected. The checklist should follow the references and follow the (optional) supplemental material. The checklist does NOT count towards the page limit. Please read the checklist guidelines carefully for information on how to answer these questions. For each question in the checklist: - You should answer [Yes] , [No] , or [N/A] . - [N/A] means either that the question is Not Applicable for that particular paper or the relevant information is Not Available. - Please provide a short (1–2 sentence) justification right after your answer (even for NA). The checklist answers are an integral part of your paper submission. They are visible to the reviewers, area chairs, senior area chairs, and ethics reviewers. You will be asked to also include it (after eventual revisions) with the final version of your paper, and its final version will be published with the paper. The reviewers of your paper will be asked to use the checklist as one of the factors in their evaluation. While " [Yes] " is generally preferable to " [No] ", it is perfectly acceptable to answer " [No] " provided a proper justification is given (e.g., "error bars are not reported because it would be too computationally expensive" or "we were unable to find the license for the dataset we used"). In general, answering " [No] " or " [N/A] " is not grounds for rejection. While the questions are phrased in a binary way, we acknowledge that the true answer is often more nuanced, so please just use your best judgment and write a justification to elaborate. All supporting evidence can appear either in the main paper or the supplemental material, provided in appendix. If you answer [Yes] to a question, in the justification please point to the section(s) where related material for the question can be found. IMPORTANT, please: - Delete this instruction block, but keep the section heading “NeurIPS Paper Checklist", - Keep the checklist subsection headings, questions/answers and guidelines below. - Do not modify the questions and only use the provided macros for your answers. 1. Claims 1. Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? 1. Answer: [Yes] 1. Justification: The abstract and the introduction summarizes our main contributions. 1. Guidelines: - The answer NA means that the abstract and introduction do not include the claims made in the paper. - The abstract and/or introduction should clearly state the claims made, including the contributions made in the paper and important assumptions and limitations. A No or NA answer to this question will not be perceived well by the reviewers. - The claims made should match theoretical and experimental results, and reflect how much the results can be expected to generalize to other settings. - It is fine to include aspirational goals as motivation as long as it is clear that these goals are not attained by the paper. 1. Limitations 1. Question: Does the paper discuss the limitations of the work performed by the authors? 1. Answer: [Yes] 1. Justification: This paper cover a section of the limiations. 1. Guidelines: - The answer NA means that the paper has no limitation while the answer No means that the paper has limitations, but those are not discussed in the paper. - The authors are encouraged to create a separate "Limitations" section in their paper. - The paper should point out any strong assumptions and how robust the results are to violations of these assumptions (e.g., independence assumptions, noiseless settings, model well-specification, asymptotic approximations only holding locally). The authors should reflect on how these assumptions might be violated in practice and what the implications would be. - The authors should reflect on the scope of the claims made, e.g., if the approach was only tested on a few datasets or with a few runs. In general, empirical results often depend on implicit assumptions, which should be articulated. - The authors should reflect on the factors that influence the performance of the approach. For example, a facial recognition algorithm may perform poorly when image resolution is low or images are taken in low lighting. Or a speech-to-text system might not be used reliably to provide closed captions for online lectures because it fails to handle technical jargon. - The authors should discuss the computational efficiency of the proposed algorithms and how they scale with dataset size. - If applicable, the authors should discuss possible limitations of their approach to address problems of privacy and fairness. - While the authors might fear that complete honesty about limitations might be used by reviewers as grounds for rejection, a worse outcome might be that reviewers discover limitations that aren’t acknowledged in the paper. The authors should use their best judgment and recognize that individual actions in favor of transparency play an important role in developing norms that preserve the integrity of the community. Reviewers will be specifically instructed to not penalize honesty concerning limitations. 1. Theory assumptions and proofs 1. Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the paper does not include theoretical results. - All the theorems, formulas, and proofs in the paper should be numbered and cross-referenced. - All assumptions should be clearly stated or referenced in the statement of any theorems. - The proofs can either appear in the main paper or the supplemental material, but if they appear in the supplemental material, the authors are encouraged to provide a short proof sketch to provide intuition. - Inversely, any informal proof provided in the core of the paper should be complemented by formal proofs provided in appendix or supplemental material. - Theorems and Lemmas that the proof relies upon should be properly referenced. 1. Experimental result reproducibility 1. Question: Does the paper fully disclose all the information needed to reproduce the main experimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and data are provided or not)? 1. Answer: [Yes] 1. Justification: Both code and datasets are available. 1. Guidelines: - The answer NA means that the paper does not include experiments. - If the paper includes experiments, a No answer to this question will not be perceived well by the reviewers: Making the paper reproducible is important, regardless of whether the code and data are provided or not. - If the contribution is a dataset and/or model, the authors should describe the steps taken to make their results reproducible or verifiable. - Depending on the contribution, reproducibility can be accomplished in various ways. For example, if the contribution is a novel architecture, describing the architecture fully might suffice, or if the contribution is a specific model and empirical evaluation, it may be necessary to either make it possible for others to replicate the model with the same dataset, or provide access to the model. In general. releasing code and data is often one good way to accomplish this, but reproducibility can also be provided via detailed instructions for how to replicate the results, access to a hosted model (e.g., in the case of a large language model), releasing of a model checkpoint, or other means that are appropriate to the research performed. - While NeurIPS does not require releasing code, the conference does require all submissions to provide some reasonable avenue for reproducibility, which may depend on the nature of the contribution. For example 1. If the contribution is primarily a new algorithm, the paper should make it clear how to reproduce that algorithm. 1. If the contribution is primarily a new model architecture, the paper should describe the architecture clearly and fully. 1. If the contribution is a new model (e.g., a large language model), then there should either be a way to access this model for reproducing the results or a way to reproduce the model (e.g., with an open-source dataset or instructions for how to construct the dataset). 1. We recognize that reproducibility may be tricky in some cases, in which case authors are welcome to describe the particular way they provide for reproducibility. In the case of closed-source models, it may be that access to the model is limited in some way (e.g., to registered users), but it should be possible for other researchers to have some path to reproducing or verifying the results. 1. Open access to data and code 1. Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? 1. Answer: [Yes] 1. Justification: We provide the code link in the abstract. 1. Guidelines: - The answer NA means that paper does not include experiments requiring code. - Please see the NeurIPS code and data submission guidelines (https://nips.cc/public/guides/CodeSubmissionPolicy) for more details. - While we encourage the release of code and data, we understand that this might not be possible, so “No” is an acceptable answer. Papers cannot be rejected simply for not including code, unless this is central to the contribution (e.g., for a new open-source benchmark). - The instructions should contain the exact command and environment needed to run to reproduce the results. See the NeurIPS code and data submission guidelines (https://nips.cc/public/guides/CodeSubmissionPolicy) for more details. - The authors should provide instructions on data access and preparation, including how to access the raw data, preprocessed data, intermediate data, and generated data, etc. - The authors should provide scripts to reproduce all experimental results for the new proposed method and baselines. If only a subset of experiments are reproducible, they should state which ones are omitted from the script and why. - At submission time, to preserve anonymity, the authors should release anonymized versions (if applicable). - Providing as much information as possible in supplemental material (appended to the paper) is recommended, but including URLs to data and code is permitted. 1. Experimental setting/details 1. Question: Does the paper specify all the training and test details (e.g., data splits, hyperparameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? 1. Answer: [Yes] 1. Justification: We cover all the details in the paper. 1. Guidelines: - The answer NA means that the paper does not include experiments. - The experimental setting should be presented in the core of the paper to a level of detail that is necessary to appreciate the results and make sense of them. - The full details can be provided either with the code, in appendix, or as supplemental material. 1. Experiment statistical significance 1. Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? 1. Answer: [No] 1. Justification: The experiments utilize the API of Large Language Models. Multiple calls will significantly increase costs. 1. Guidelines: - The answer NA means that the paper does not include experiments. - The authors should answer "Yes" if the results are accompanied by error bars, confidence intervals, or statistical significance tests, at least for the experiments that support the main claims of the paper. - The factors of variability that the error bars are capturing should be clearly stated (for example, train/test split, initialization, random drawing of some parameter, or overall run with given experimental conditions). - The method for calculating the error bars should be explained (closed form formula, call to a library function, bootstrap, etc.) - The assumptions made should be given (e.g., Normally distributed errors). - It should be clear whether the error bar is the standard deviation or the standard error of the mean. - It is OK to report 1-sigma error bars, but one should state it. The authors should preferably report a 2-sigma error bar than state that they have a 96% CI, if the hypothesis of Normality of errors is not verified. - For asymmetric distributions, the authors should be careful not to show in tables or figures symmetric error bars that would yield results that are out of range (e.g. negative error rates). - If error bars are reported in tables or plots, The authors should explain in the text how they were calculated and reference the corresponding figures or tables in the text. 1. Experiments compute resources 1. Question: For each experiment, does the paper provide sufficient information on the computer resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? 1. Answer: [Yes] 1. Justification: It could be found in the experimental part. 1. Guidelines: - The answer NA means that the paper does not include experiments. - The paper should indicate the type of compute workers CPU or GPU, internal cluster, or cloud provider, including relevant memory and storage. - The paper should provide the amount of compute required for each of the individual experimental runs as well as estimate the total compute. - The paper should disclose whether the full research project required more compute than the experiments reported in the paper (e.g., preliminary or failed experiments that didn’t make it into the paper). 1. Code of ethics 1. Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the authors have not reviewed the NeurIPS Code of Ethics. - If the authors answer No, they should explain the special circumstances that require a deviation from the Code of Ethics. - The authors should make sure to preserve anonymity (e.g., if there is a special consideration due to laws or regulations in their jurisdiction). 1. Broader impacts 1. Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? 1. Answer: [No] 1. Justification: We don’t discuss this aspect because we provide only the memory system for LLM agents. Different LLM agents may create varying societal impacts, which are beyond the scope of our work. 1. Guidelines: - The answer NA means that there is no societal impact of the work performed. - If the authors answer NA or No, they should explain why their work has no societal impact or why the paper does not address societal impact. - Examples of negative societal impacts include potential malicious or unintended uses (e.g., disinformation, generating fake profiles, surveillance), fairness considerations (e.g., deployment of technologies that could make decisions that unfairly impact specific groups), privacy considerations, and security considerations. - The conference expects that many papers will be foundational research and not tied to particular applications, let alone deployments. However, if there is a direct path to any negative applications, the authors should point it out. For example, it is legitimate to point out that an improvement in the quality of generative models could be used to generate deepfakes for disinformation. On the other hand, it is not needed to point out that a generic algorithm for optimizing neural networks could enable people to train models that generate Deepfakes faster. - The authors should consider possible harms that could arise when the technology is being used as intended and functioning correctly, harms that could arise when the technology is being used as intended but gives incorrect results, and harms following from (intentional or unintentional) misuse of the technology. - If there are negative societal impacts, the authors could also discuss possible mitigation strategies (e.g., gated release of models, providing defenses in addition to attacks, mechanisms for monitoring misuse, mechanisms to monitor how a system learns from feedback over time, improving the efficiency and accessibility of ML). 1. Safeguards 1. Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the paper poses no such risks. - Released models that have a high risk for misuse or dual-use should be released with necessary safeguards to allow for controlled use of the model, for example by requiring that users adhere to usage guidelines or restrictions to access the model or implementing safety filters. - Datasets that have been scraped from the Internet could pose safety risks. The authors should describe how they avoided releasing unsafe images. - We recognize that providing effective safeguards is challenging, and many papers do not require this, but we encourage authors to take this into account and make a best faith effort. 1. Licenses for existing assets 1. Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? 1. Answer: [Yes] 1. Justification: Their contribution has already been properly acknowledged and credited. 1. Guidelines: - The answer NA means that the paper does not use existing assets. - The authors should cite the original paper that produced the code package or dataset. - The authors should state which version of the asset is used and, if possible, include a URL. - The name of the license (e.g., CC-BY 4.0) should be included for each asset. - For scraped data from a particular source (e.g., website), the copyright and terms of service of that source should be provided. - If assets are released, the license, copyright information, and terms of use in the package should be provided. For popular datasets, paperswithcode.com/datasets has curated licenses for some datasets. Their licensing guide can help determine the license of a dataset. - For existing datasets that are re-packaged, both the original license and the license of the derived asset (if it has changed) should be provided. - If this information is not available online, the authors are encouraged to reach out to the asset’s creators. 1. New assets 1. Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the paper does not release new assets. - Researchers should communicate the details of the dataset/code/model as part of their submissions via structured templates. This includes details about training, license, limitations, etc. - The paper should discuss whether and how consent was obtained from people whose asset is used. - At submission time, remember to anonymize your assets (if applicable). You can either create an anonymized URL or include an anonymized zip file. 1. Crowdsourcing and research with human subjects 1. Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the paper does not involve crowdsourcing nor research with human subjects. - Including this information in the supplemental material is fine, but if the main contribution of the paper involves human subjects, then as much detail as possible should be included in the main paper. - According to the NeurIPS Code of Ethics, workers involved in data collection, curation, or other labor should be paid at least the minimum wage in the country of the data collector. 1. Institutional review board (IRB) approvals or equivalent for research with human subjects 1. Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the requirements of your country or institution) were obtained? 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the paper does not involve crowdsourcing nor research with human subjects. - Depending on the country in which research is conducted, IRB approval (or equivalent) may be required for any human subjects research. If you obtained IRB approval, you should clearly state this in the paper. - We recognize that the procedures for this may vary significantly between institutions and locations, and we expect authors to adhere to the NeurIPS Code of Ethics and the guidelines for their institution. - For initial submissions, do not include any information that would break anonymity (if applicable), such as the institution conducting the review. 1. Declaration of LLM usage 1. Question: Does the paper describe the usage of LLMs if it is an important, original, or non-standard component of the core methods in this research? Note that if the LLM is used only for writing, editing, or formatting purposes and does not impact the core methodology, scientific rigorousness, or originality of the research, declaration is not required. 1. Answer: [N/A] 1. Justification: N/A 1. Guidelines: - The answer NA means that the core method development in this research does not involve LLMs as any important, original, or non-standard components. - Please refer to our LLM policy (https://neurips.cc/Conferences/2025/LLM) for what should or should not be described.