{
  "reviewed": "2026-10-17",
  "attribution": "Company-reported experimental results and reference methods transcribed from the latest Memory comparison table and the earlier consolidated evaluation report.",
  "configuration": "Results come from multiple source reports. Older rows include paper-reported references and company runs; matching models, samples and settings are not established. Compare only within the stated metric and subset; these are not a controlled cross-system reproduction.",
  "systems": [
    {
      "id": "cosens",
      "name": "CoSens",
      "sourceLabel": "Friday",
      "description": "Experience layer with persistent memory and reusable methods. The reported memory implementation is named Friday in the source reports.",
      "links": [
        {
          "title": "Website",
          "url": "https://cosens.ai/"
        },
        {
          "title": "SDK preview",
          "url": "https://cosens.ai/docs/#overview"
        }
      ],
      "evaluationNote": "Company-run memory results. These scores do not evaluate the complete SDK or the Learn loop; a public evaluated source revision is not supplied."
    },
    {
      "id": "mem0",
      "name": "Mem0",
      "sourceLabel": "Mem0",
      "description": "A memory layer that extracts, stores and retrieves information for agents and applications.",
      "links": [
        {
          "title": "Official repository",
          "url": "https://github.com/mem0ai/mem0"
        }
      ],
      "evaluationNote": "Latest-report rows use company-run Mem0 results; MemoryAgentBench explicitly uses Mem0 OSS. LoCoMo-Plus uses a paper reference. Platform, OSS and graph variants must not be treated as interchangeable."
    },
    {
      "id": "tencent",
      "name": "TencentDB Agent Memory",
      "sourceLabel": "TencentDB Agent Memory",
      "description": "TencentDB Agent Memory provides persistent memory infrastructure for agents.",
      "links": [
        {
          "title": "Official repository",
          "url": "https://github.com/TencentCloud/TencentDB-Agent-Memory"
        }
      ],
      "evaluationNote": "Company-run results in the latest report. The exact evaluated release and storage configuration are not recorded; the current repository may differ from the evaluated version."
    },
    {
      "id": "amem",
      "name": "A-Mem",
      "sourceLabel": "A-Mem",
      "description": "Agentic memory that organizes notes, metadata and links, and updates relationships as new memories arrive.",
      "links": [
        {
          "title": "Official memory implementation",
          "url": "https://github.com/agiresearch/A-mem"
        },
        {
          "title": "Authors’ evaluation code",
          "url": "https://github.com/WujiangXu/AgenticMemory"
        },
        {
          "title": "Paper",
          "url": "https://arxiv.org/abs/2502.12110"
        }
      ],
      "evaluationNote": "LoCoMo-Plus values are paper references reproduced in the company report. The system repository distinguishes the memory implementation from the authors’ experiment code."
    },
    {
      "id": "secom",
      "name": "SeCom",
      "sourceLabel": "SeCom",
      "description": "Conversation memory built from topical segments, with compression and retrieval over those memory units.",
      "links": [
        {
          "title": "Official repository and experiments",
          "url": "https://github.com/microsoft/SeCom"
        }
      ],
      "evaluationNote": "LoCoMo-Plus values are paper references reproduced in the company report; they are not a new controlled run of the current SeCom release."
    },
    {
      "id": "memgpt",
      "name": "MemGPT",
      "sourceLabel": "MemGPT",
      "description": "The MemGPT agent design manages a limited context window using external memory and memory-editing tools.",
      "links": [
        {
          "title": "Project repository · now Letta",
          "url": "https://github.com/letta-ai/letta"
        },
        {
          "title": "Original MemGPT paper",
          "url": "https://arxiv.org/abs/2310.08560"
        }
      ],
      "evaluationNote": "MemBench uses the historical MemGPT reference from the source report. The current Letta project is its successor, not proof that this score measures today’s Letta release."
    },
    {
      "id": "raglocal",
      "name": "RAG (local)",
      "sourceLabel": "ragmemory",
      "description": "The local retrieval-augmented memory baseline labeled ragmemory in the consolidated report.",
      "links": [
        {
          "title": "Baseline source report",
          "url": "https://alidocs.dingtalk.com/i/nodes/93NwLYZXWyEw0OPZiNEg9mBYWkyEqBQm"
        }
      ],
      "evaluationNote": "A company-run baseline, not a separate publicly identified product. The report describes matched local conditions for its MemBench comparison but supplies no public implementation URL or complete configuration."
    },
    {
      "id": "sessionrag",
      "name": "Session RAG (V+fact)",
      "sourceLabel": "Session / K=V+fact",
      "description": "A LongMemEval paper reference using sessions as retrieval values, with session text and extracted facts as retrieval keys (K=V+fact).",
      "links": [
        {
          "title": "Authors’ code and data",
          "url": "https://github.com/xiaowu0162/LongMemEval"
        },
        {
          "title": "Paper · retrieval design comparison",
          "url": "https://arxiv.org/abs/2410.10813"
        }
      ],
      "evaluationNote": "The 86.2 value is the paper’s Table 3 retrieval reference transcribed into the company report. It is a method configuration, not an independently released memory product."
    }
  ],
  "benchmarks": [
    {
      "name": "LoCoMo",
      "url": "https://github.com/snap-research/LoCoMo",
      "metric": "Overall Token-level F1",
      "scale": "0–1; higher is better",
      "scores": {
        "cosens": 0.368945,
        "mem0": 0.256981,
        "tencent": 0.186029,
        "amem": null,
        "secom": null,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": null
      },
      "source": "Latest Memory comparison table",
      "precision": 6,
      "id": "locomo",
      "sourceId": "latest",
      "links": [
        {
          "title": "Official code and dataset",
          "url": "https://github.com/snap-research/LoCoMo"
        },
        {
          "title": "Paper",
          "url": "https://arxiv.org/abs/2402.17753"
        }
      ],
      "methodology": {
        "purpose": "Tests whether an assistant can answer questions using information distributed across a long conversation history.",
        "dataset": "The public release contains ten multi-session conversations with timestamps, QA labels and evidence dialogue IDs.",
        "procedure": "Ingest conversation history, retrieve relevant memory for each question, generate an answer, then compare it with the annotated answer using the QA evaluator.",
        "scoring": "This row reports Overall Token-level F1 on a 0–1 scale: token precision and recall measure overlap with the reference answer. Higher is better; no LLM judge is needed for this selected metric.",
        "scope": "All three populated cells come from the latest company report. Sample selection, category inclusion, answer normalization and the exact Overall aggregation are not supplied; the repository’s full dataset size does not establish the company run size."
      },
      "scoreOrigins": {
        "cosens": "company-run",
        "mem0": "company-run",
        "tencent": "company-run"
      }
    },
    {
      "name": "HaluMem",
      "metric": "Medium QA Correct",
      "scale": "0–1; higher is better",
      "scores": {
        "cosens": 0.537369,
        "mem0": 0.492279,
        "tencent": 0.519456,
        "amem": null,
        "secom": null,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": null
      },
      "source": "Latest Memory comparison table",
      "precision": 6,
      "id": "halumem",
      "sourceId": "latest",
      "links": [
        {
          "title": "Official code and evaluation suite",
          "url": "https://github.com/MemTensor/HaluMem"
        },
        {
          "title": "Paper",
          "url": "https://arxiv.org/abs/2511.03506"
        },
        {
          "title": "Official dataset",
          "url": "https://huggingface.co/datasets/IAAR-Shanghai/HaluMem"
        }
      ],
      "methodology": {
        "purpose": "Tests hallucinations in memory extraction, memory updating and downstream question answering as separate operations.",
        "dataset": "HaluMem offers Medium and Long user-centric conversation histories, annotated memory points and QA pairs. This row selects Medium QA only.",
        "procedure": "Feed sessions through a memory-system adapter, collect stored and updated memories and QA predictions, then run the evaluation suite against annotated reference information.",
        "scoring": "Medium QA Correct is the reported correctness measure on a 0–1 scale. The official suite also evaluates memory extraction and update accuracy, hallucinations and omissions; those are not averaged into this row.",
        "scope": "The three values are company-run results in the latest report. Judge version, prompts, thresholds and evaluated QA count are not recorded. A high QA value alone does not establish extraction or update leadership."
      },
      "url": "https://github.com/MemTensor/HaluMem",
      "scoreOrigins": {
        "cosens": "company-run",
        "mem0": "company-run",
        "tencent": "company-run"
      }
    },
    {
      "name": "MemoryAgentBench",
      "metric": "Conflict Resolution · Token F1",
      "scale": "0–1; higher is better",
      "sourceLabels": {
        "mem0": "Mem0 OSS"
      },
      "scores": {
        "cosens": null,
        "mem0": 0.035893,
        "tencent": 0.038262,
        "amem": null,
        "secom": null,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": null
      },
      "source": "Latest Memory comparison table",
      "precision": 6,
      "id": "memoryagentbench",
      "sourceId": "latest",
      "links": [
        {
          "title": "Official code and task definitions",
          "url": "https://github.com/HUST-AI-HYZ/MemoryAgentBench"
        },
        {
          "title": "Paper",
          "url": "https://arxiv.org/abs/2507.05257"
        }
      ],
      "methodology": {
        "purpose": "Tests memory through incremental interactions, including accurate retrieval, test-time learning, long-range understanding and conflict resolution.",
        "dataset": "The selected Conflict Resolution task provides facts that change or conflict across incoming context chunks.",
        "procedure": "Process chunks into memory in sequence, then answer questions that require the applicable updated fact rather than an outdated conflicting version.",
        "scoring": "MemoryAgentBench uses an adapted Token F1 in this company report, on a 0–1 scale. The official repository describes substring exact-match accuracy for its conflict-resolution tasks. These scoring protocols are different.",
        "scope": "Only Mem0 OSS and TencentDB have reported values; the CoSens result is missing. The adaptation code, chunk size, task revision, sample count and normalization rules are not supplied, so these values are not official-protocol accuracy."
      },
      "url": "https://github.com/HUST-AI-HYZ/MemoryAgentBench",
      "scoreOrigins": {
        "mem0": "company-run",
        "tencent": "company-run"
      }
    },
    {
      "name": "OrgMemBench",
      "metric": "Medium · Overall rubric score",
      "scale": "higher is better",
      "scores": {
        "cosens": 0.3241,
        "mem0": null,
        "tencent": null,
        "amem": null,
        "secom": null,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": null
      },
      "source": "Latest Memory comparison table",
      "precision": 6,
      "id": "orgmembench",
      "sourceId": "latest",
      "links": [
        {
          "title": "Official corpus and harness",
          "url": "https://github.com/JackCGardner/OrgMemBench"
        },
        {
          "title": "Methodology",
          "url": "https://github.com/JackCGardner/OrgMemBench/blob/main/docs/METHODOLOGY.md"
        },
        {
          "title": "Reproduction guide",
          "url": "https://github.com/JackCGardner/OrgMemBench/blob/main/docs/REPRODUCIBILITY.md"
        }
      ],
      "methodology": {
        "purpose": "Tests organizational memory: revised decisions, provenance, historical belief states, audit replay, justification chains and contradictions.",
        "dataset": "A synthetic organization with multiple authors, channels and valid-time versus ingestion-time history. The public Medium tier has 443 artifacts and 73 questions.",
        "procedure": "Ingest the selected tier with a system adapter, retrieve evidence, generate answers and judge them against structured rubric facets. The harness supports both native and shared answerers.",
        "scoring": "Overall rubric score is mean rubric-weighted facet coverage in [0,1], allowing partial credit. This row keeps the company report’s Medium result; its Small result is not mixed in.",
        "scope": "Only the CoSens company-run value is reported here. The public tier size describes the benchmark; actual evaluated question IDs, corpus revision, answerer, judge and run count are not supplied by the company report."
      },
      "url": "https://github.com/JackCGardner/OrgMemBench",
      "scoreOrigins": {
        "cosens": "company-run"
      }
    },
    {
      "name": "LongMemEval",
      "metric": "Evidence Recall@10 (%)",
      "scale": "0–100; higher is better",
      "source": "Earlier consolidated report · retrieval evidence, not answer accuracy",
      "scores": {
        "cosens": 95.2,
        "mem0": null,
        "tencent": null,
        "amem": null,
        "secom": null,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": 86.2
      },
      "precision": 1,
      "id": "longmemeval",
      "sourceId": "earlier",
      "links": [
        {
          "title": "Official code, data and evaluation",
          "url": "https://github.com/xiaowu0162/LongMemEval"
        },
        {
          "title": "Paper · Table 3",
          "url": "https://arxiv.org/abs/2410.10813"
        }
      ],
      "methodology": {
        "purpose": "Tests long-term interactive memory, including extraction, multi-session and temporal reasoning, knowledge updates and abstention.",
        "dataset": "Public S, M and oracle variants each contain 500 instances, with evidence session IDs and answer-bearing turn annotations. The company report does not identify its exact variant.",
        "procedure": "For retrieval evaluation, index the history, retrieve the ten highest-ranked memory items for each question, and check against annotated evidence. The benchmark separately supports generated-answer evaluation.",
        "scoring": "Evidence Recall@10 (%) measures reference evidence recovered in the top ten results, on a 0–100 scale. It is not end-to-end answer accuracy. The source does not provide the exact matching and aggregation implementation.",
        "scope": "CoSens is a company-run result; Session RAG (V+fact) is a Table 3 paper reference. Retrieval unit, evidence matching, embedding version, query expansion and sample selection are not established as matched."
      },
      "url": "https://github.com/xiaowu0162/LongMemEval",
      "scoreOrigins": {
        "cosens": "company-run",
        "sessionrag": "paper-reference"
      }
    },
    {
      "name": "LoCoMo-Plus",
      "metric": "Factual average",
      "scale": "0–100; higher is better",
      "source": "Earlier consolidated report · paper references and company result",
      "scores": {
        "cosens": 61.52,
        "mem0": 57.24,
        "tencent": null,
        "amem": 59.64,
        "secom": 57.53,
        "memgpt": null,
        "raglocal": null,
        "sessionrag": null
      },
      "precision": 2,
      "id": "locomo-plus",
      "sourceId": "earlier",
      "links": [
        {
          "title": "Official code, data and judge framework",
          "url": "https://github.com/xjtuleeyf/Locomo-Plus"
        },
        {
          "title": "Paper",
          "url": "https://arxiv.org/abs/2602.10715"
        }
      ],
      "methodology": {
        "purpose": "Adds cognitive memory to long conversational QA: use an earlier cue to respect a later implicit constraint, even when their wording differs.",
        "dataset": "Extends LoCoMo’s factual categories with cue–trigger cognitive tasks. This table selects the reported factual average, not the cognitive score or a combined total.",
        "procedure": "Build unified inputs, obtain predictions from the model or memory system, then score with the released LLM-as-judge framework.",
        "scoring": "The released judge awards correct, partial and wrong responses 1, 0.5 and 0. This row preserves the report’s factual average on a 0–100 scale; the company aggregation implementation is not supplied.",
        "scope": "CoSens is a company-run value. Mem0, A-Mem and SeCom are paper references copied into the consolidated report. Model, prompt, judge and exact sample equivalence have not been established."
      },
      "url": "https://github.com/xjtuleeyf/Locomo-Plus",
      "scoreOrigins": {
        "cosens": "company-run",
        "mem0": "paper-reference",
        "amem": "paper-reference",
        "secom": "paper-reference"
      }
    },
    {
      "name": "MemBench",
      "metric": "Participation Scenario · Accuracy",
      "scale": "0–1; higher is better",
      "source": "Earlier consolidated report · paper references and local baseline",
      "scores": {
        "cosens": 0.892,
        "mem0": null,
        "tencent": null,
        "amem": null,
        "secom": null,
        "memgpt": 0.455,
        "raglocal": 0.722,
        "sessionrag": null
      },
      "precision": 3,
      "id": "membench",
      "sourceId": "earlier",
      "links": [
        {
          "title": "Official repository and data instructions",
          "url": "https://github.com/import-myself/Membench"
        },
        {
          "title": "Paper",
          "url": "https://aclanthology.org/2025.findings-acl.989/"
        }
      ],
      "methodology": {
        "purpose": "Tests factual and reflective agent memory in participation and observation scenarios, with effectiveness, efficiency and capacity dimensions.",
        "dataset": "Participation data represents first-person interaction; observation data represents third-person information. Released data supports different conversation lengths and added distractor content.",
        "procedure": "Feed the chosen interaction history through a memory method, answer the associated questions and score correctness for the selected scenario and context-length condition.",
        "scoring": "This row uses reported Participation Scenario Accuracy on a 0–1 scale. Observation accuracy, efficiency and capacity are separate measures and are not combined into this value.",
        "scope": "CoSens and RAG (local) are company-run results; MemGPT is a paper reference. The report describes its local RAG comparison as matched, but exact lengths, question IDs, sample counts and scoring settings are not supplied."
      },
      "url": "https://github.com/import-myself/Membench",
      "scoreOrigins": {
        "cosens": "company-run",
        "memgpt": "paper-reference",
        "raglocal": "local-baseline"
      }
    }
  ],
  "omitted": [],
  "selection": "Use reported Overall scores where available; otherwise use the task-facing QA or answer-quality metric. Preserve the task and subset label. Missing results are null, displayed as —.",
  "disclaimer": "Experimental results are provided for reference only and do not constitute academic research. Results may be updated at any time as new evaluations become available.",
  "homepage": {
    "systemIds": [
      "cosens",
      "mem0",
      "amem",
      "memgpt"
    ],
    "benchmarkNames": [
      "LoCoMo",
      "LoCoMo-Plus",
      "HaluMem",
      "LongMemEval"
    ]
  },
  "sources": {
    "latest": {
      "title": "Latest Memory comparison table",
      "url": "https://alidocs.dingtalk.com/i/nodes/oP0MALyR8kENMeLKiMgGXZqBV3bzYmDO",
      "access": "Company report · DingTalk permission required",
      "description": "Source of the LoCoMo, HaluMem, MemoryAgentBench and OrgMemBench scores. Reviewed 17 October 2026."
    },
    "earlier": {
      "title": "Earlier consolidated Memory evaluation report",
      "url": "https://alidocs.dingtalk.com/i/nodes/93NwLYZXWyEw0OPZiNEg9mBYWkyEqBQm",
      "access": "Company report · DingTalk permission required",
      "description": "Source of the LongMemEval, LoCoMo-Plus and MemBench rows. Combines company results, published references and a local RAG baseline. Reviewed 17 October 2026."
    }
  },
  "runMetadata": [
    {
      "field": "Dataset revision and split",
      "status": "Not provided",
      "detail": "No pinned dataset commit, checksum or complete split declaration accompanies the reported runs."
    },
    {
      "field": "Sample IDs and counts",
      "status": "Not provided",
      "detail": "Public dataset sizes are described below; actual run manifests and evaluated counts are not supplied."
    },
    {
      "field": "Memory implementation versions",
      "status": "Not provided",
      "detail": "No evaluated commit, package version or deployment revision is supplied for the company runs."
    },
    {
      "field": "Generator and embedding models",
      "status": "Not provided",
      "detail": "No complete per-system model versions, embedding configuration or provider settings are supplied."
    },
    {
      "field": "Retrieval and ingestion settings",
      "status": "Partly provided",
      "detail": "LongMemEval selects top 10; other chunking, top-k, filtering, memory-write and update settings are not supplied."
    },
    {
      "field": "Judge, prompts and aggregation",
      "status": "Partly provided",
      "detail": "Selected metrics are named. Exact company evaluator revisions, prompts, normalization and aggregation code are not supplied."
    },
    {
      "field": "Run dates, repetitions and uncertainty",
      "status": "Not provided",
      "detail": "Report review date is not a run date. Seeds, repeat counts and confidence intervals are not supplied."
    },
    {
      "field": "Raw predictions and run artifacts",
      "status": "Not provided",
      "detail": "This page publishes transcribed scores and references; raw per-question outputs, execution logs and reproduction commands for the company runs are not supplied."
    }
  ]
}