[{"data":1,"prerenderedAt":257},["ShallowReactive",2],{"docs-page-en-/usecase/blog/million_token_context_long_term_memory":3,"surround-en-/usecase/blog/million_token_context_long_term_memory":245},{"id":4,"title":5,"avatar":6,"banner":6,"body":7,"category":6,"desc":6,"description":231,"extension":239,"lead":6,"links":6,"meta":240,"navigation":6,"path":241,"seo":242,"stem":243,"__hash__":244},"docs/en/usecase/blog/million_token_context_long_term_memory.md","Why Long-Term Memory Still Matters with a Million-Token Context Window",null,{"type":8,"value":9,"toc":230},"minimark",[10,21,24,27,30,33,38,41,108,111,113,117,120,123,126,129,131,135,146,151,160,165,186,191,200,202,206,209,212,215,217,221,224,227],[11,12,13],"blockquote",{},[14,15,16,20],"p",{},[17,18,19],"strong",{},"Key takeaway:"," Users want an AI assistant to understand a long document and remember arrangements confirmed last time. They should not have to explain a budget again in a new session; recommendations should adapt when preferences change; and information they no longer want retained should be deletable.",[22,23],"hr",{},[14,25,26],{},"A large context window expands how much information a model can process at once. Long-term memory concerns how that information carries into the next interaction: what deserves to be retained, whom it belongs to, when it changes, and when to stop using it.",[14,28,29],{},"At MemTensor, we use MemOS to manage this process, helping agents accumulate useful information through ongoing interactions and retrieve it in later tasks. These cross-session management needs remain even with a million-token context window.",[31,32],"br",{},[34,35,37],"h2",{"id":36},"how-do-the-four-capabilities-work-together","How do the four capabilities work together?",[14,39,40],{},"Context windows, RAG, vector databases, and long-term memory operate at different layers and can serve the same task together.",[42,43,44,60],"table",{},[45,46,47],"thead",{},[48,49,50,54,57],"tr",{},[51,52,53],"th",{},"Capability",[51,55,56],{},"Role",[51,58,59],{},"Use in customer support",[61,62,63,75,86,97],"tbody",{},[48,64,65,69,72],{},[66,67,68],"td",{},"Context window",[66,70,71],{},"Holds the information a model needs for the current request",[66,73,74],{},"Contains the current question, recent conversation, and retrieval results",[48,76,77,80,83],{},[66,78,79],{},"RAG",[66,81,82],{},"Retrieves external material to ground an answer",[66,84,85],{},"Finds product manuals and warranty policies",[48,87,88,91,94],{},[66,89,90],{},"Vector database",[66,92,93],{},"Stores vector representations and supports semantic similarity search",[66,95,96],{},"Supplies candidate documents or memories",[48,98,99,102,105],{},[66,100,101],{},"Long-term memory system",[66,103,104],{},"Manages storage, use, updates, and forgetting of facts and experience across sessions",[66,106,107],{},"Continues from confirmed customer preferences and service progress",[14,109,110],{},"RAG can use vector databases alongside keyword, graph, and other retrieval methods. Long-term memory can use these capabilities too, bringing information suitable for the current task into context. The model handles understanding and reasoning; the memory system continually maintains usable information.",[31,112],{},[34,114,116],{"id":115},"follow-a-user-fact-as-it-changes","Follow a user fact as it changes",[14,118,119],{},"On Monday, a user tells a travel assistant: “My nightly budget for this Shanghai trip is RMB 800.” On Wednesday, she adds: “The company will reimburse it, so raise the limit to RMB 1,200, but stay close to the venue.” On Friday, she opens a new session to continue choosing a hotel.",[14,121,122],{},"The assistant needs to resume that trip with the latest budget and location requirements, without asking her to repeat the earlier conversations. If she asks about the initial plan, it should also be able to find the historical RMB 800 budget.",[14,124,125],{},"Scope matters too: RMB 1,200 is the budget for this business trip, rather than a default for every future trip. Memories need task, time, and source information so later retrieval can select applicable facts.",[14,127,128],{},"A larger context window gives the model room to understand the full history. Long-term memory lets the application maintain the trip over time and bring back the relevant details when the user returns.",[31,130],{},[34,132,134],{"id":133},"how-memos-connects-the-steps","How MemOS connects the steps",[14,136,137,138,145],{},"With ",[139,140,144],"a",{"href":141,"rel":142},"https://memos-docs.openmem.net/memos_cloud/getting_started/overview/",[143],"nofollow","MemOS Cloud",", developers submit conversation messages for processing and storage as memories, then retrieve them in subsequent requests. The travel example involves three steps.",[14,147,148],{},[17,149,150],{},"Organize information by purpose.",[14,152,153,154,159],{},"MemOS supports ",[139,155,158],{"href":156,"rel":157},"https://memos-docs.openmem.net/memos_cloud/introduction/memory_types/",[143],"fact, preference, profile, event, skill, tool, and knowledge base memories",". Facts and events can capture trip requirements and arrangements; preferences help personalize recommendations; knowledge bases can supply the company's travel policy. During retrieval, applications can select memory types and combine user memories with knowledge base content to provide a fuller basis for hotel recommendations.",[14,161,162],{},[17,163,164],{},"Keep memories aligned with changes.",[14,166,167,168,173,174,179,180,185],{},"MemOS ",[139,169,172],{"href":170,"rel":171},"https://memos-docs.openmem.net/memos_cloud/introduction/time_awareness/",[143],"time awareness"," helps retrieval distinguish current information from historical records. After the budget changes, the current task should use the new conditions; reviewing the original plan can still draw on the earlier record. Developers can ",[139,175,178],{"href":176,"rel":177},"https://memos-docs.openmem.net/memos_cloud/mem_operations/update_memory/",[143],"update a memory by ID"," or submit ",[139,181,184],{"href":182,"rel":183},"https://memos-docs.openmem.net/memos_cloud/mem_operations/add_feedback/",[143],"natural-language feedback",", such as “The budget for this trip has changed,” so the system can locate and update relevant memories.",[14,187,188],{},[17,189,190],{},"Give users control over retained information.",[14,192,193,194,199],{},"MemOS Cloud offers ",[139,195,198],{"href":196,"rel":197},"https://memos-docs.openmem.net/memos_cloud/mem_operations/delete_memory/",[143],"memory deletion"," for specific memories or a user's memories within a project. Agent-level deletion is also available when independent agent memory is enabled. Applications can connect these operations to settings or support workflows and handle their own related copies, making memory management part of the product experience.",[31,201],{},[34,203,205],{"id":204},"start-with-a-task-that-benefits-from-memory","Start with a task that benefits from memory",[14,207,208],{},"For one-off document analysis, prioritize large context and on-demand retrieval. For an ongoing travel assistant, support agent, or development assistant, long-term memory helps users continue where they left off, reduces repetition, and reuses confirmed information.",[14,210,211],{},"Choose one concrete scenario and test four actions: retrieve information across sessions, change a fact, check access boundaries between users, and delete a test memory. Observe whether subsequent answers reflect these changes to judge whether memory serves the task.",[14,213,214],{},"Context capacity determines how much information fits into one request. When users keep working with an agent, they experience whether it can continue established plans and adapt to new requirements. Long-term memory makes that continuity a product capability that can be designed and maintained.",[31,216],{},[34,218,220],{"id":219},"about-memtensor","About MemTensor",[14,222,223],{},"MemTensor (Shanghai) Technology Co., Ltd. is a next-generation large-model and long-term intelligence infrastructure company incubated by the Shanghai Institute for Algorithms Innovation, with a member of the Chinese Academy of Sciences serving as chief advisor.",[14,225,226],{},"Focused on reducing hallucinations, personalization, and self-learning and evolution, the company studies long-term memory and continual learning for large models. Its roadmap connects Memory³ research, the MemOS memory operating system, agent and memory infrastructure products, and exploration of memory-native general-purpose foundation models, progressing from theory to systems engineering and model research to advance AI beyond one-off generation toward long-term intelligence.",[14,228,229],{},"MemTensor has established close collaborations with partners including China Merchants, Haicheng, and HONOR, and commercial deployments in AI companionship, games, smart devices, finance, and industry. It has raised nearly RMB 200 million in total, with investors including CICC, Futeng Capital, Hubble, SenseTime, and MSA Capital.",{"title":231,"searchDepth":232,"depth":232,"links":233},"",2,[234,235,236,237,238],{"id":36,"depth":232,"text":37},{"id":115,"depth":232,"text":116},{"id":133,"depth":232,"text":134},{"id":204,"depth":232,"text":205},{"id":219,"depth":232,"text":220},"md",{},"/en/usecase/blog/million_token_context_long_term_memory",{"title":5,"description":231},"en/usecase/blog/million_token_context_long_term_memory","mkRYqsPTuQhIZVm621REipiyPECfNLXt93Psd-8pkuA",[246,252],{"title":247,"path":248,"stem":249,"framework":6,"module":6,"class":250,"target":-1,"active":251,"defaultOpen":251,"children":-1,"description":-1},"From Data Ownership to Deletion: Four Questions for Self-Hosted Agent Memory","/usecase/blog/self_hosted_agent_memory_four_questions","usecase/blog/self_hosted_agent_memory_four_questions",[],false,{"title":253,"path":254,"stem":255,"framework":6,"module":6,"class":256,"target":-1,"active":251,"defaultOpen":251,"children":-1,"description":-1},"Why Agents Need Memory Management Even with a Vector Database","/usecase/blog/vector_database_memory_management","usecase/blog/vector_database_memory_management",[],1793163800229]