记忆抽取有问题

[bjmajic]

Pre-submission checklist | 提交前检查

• I have searched existing issues and this hasn't been mentioned before | 我已搜索现有问题，确认此问题尚未被提及
• I have read the project documentation and confirmed this issue doesn't already exist | 我已阅读项目文档并确认此问题尚未存在
• This issue is specific to MemOS and not a general software issue | 该问题是针对 MemOS 的，而不是一般软件问题

Bug Description | 问题描述

使用的是 mode=fast
但是看落库的信息，感觉就是我对话的上下文原文，并没有经过提取

。 部署的容器也没有报错：

对应的 env文件为：# OpenAI API Key (use Bailian's API_KEY)
OPENAI_API_KEY=

OpenAI API Base URL

OPENAI_API_BASE=

Default model name

MOS_CHAT_MODEL=DeepSeek-V4-Flash

Memory Reader LLM Model

MEMRADER_MODEL=DeepSeek-V4-Flash

Memory Reader API Key (use Bailian's API_KEY)

MEMRADER_API_KEY=

Memory Reader API Base URL

MEMRADER_API_BASE=

api/?type=model&url=2846066
MOS_EMBEDDER_MODEL=bge-m3

Configure embedding backend, two options: ollama | universal_api

MOS_EMBEDDER_BACKEND=universal_api

Embedder API Base URL

MOS_EMBEDDER_API_BASE=

Embedder API Key (use Bailian's API_KEY)

MOS_EMBEDDER_API_KEY=

Embedding vector dimension

EMBEDDING_DIMENSION=1024

Reranker Backend (http_bge | etc.)

MOS_RERANKER_BACKEND=http_bge
MOS_RERANKER_URL=
MOS_RERANKER_MODEL=bge-reranker-v2-m3

Neo4j Connection URI

Available options: neo4j-community | neo4j | nebular | polardb

NEO4J_BACKEND=neo4j-community

Required when backend=neo4j*

NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j

Whether to use Redis schedule

DEFAULT_USE_REDIS_QUEUE=false

Enable Chat API

ENABLE_CHAT_API=true

① 检索模式默认值（插件会显式传 mode，此项主要兜底）

SEARCH_MODE=fast

② 如果 search_mode = fine 增强策略：rewrite（保留记忆 id 改写文本，推荐）| recreate（重建新记忆，默认值，id 会变）

FINE_STRATEGY 别选 deep_search / agentic_search：会走多轮 agent 深搜，慢且依赖额外组件。rewrite 最稳，recreate 是默认但会丢原记忆 id（>影响记忆图谱引用）。

FINE_STRATEGY=rewrite

检索策略

FAST_GRAPH 对 fine 不是必需：fine 模式下 keys/tags 由 LLM 填充（非空），即使 FAST_GRAPH=false 也会命中 key/tag 结构化查询。上面建议开它

主要是补 fulltext/BM25 路径。
FAST_GRAPH=true
BM25_CALL=true
#与 FAST_GRAPH 有功能重叠（都走 fulltext 路径），开 FAST_GRAPH 后可以先不开，效果不足再加
#FULLTEXT_CALL=true

④ 可选：若想让 LLM 解析出的 internet_search=True 真正联网

ENABLE_INTERNET=true

INTERNET_SEARCH_BACKEND=bocha

BOCHA_API_KEY=你的key

调度器开关和可选的调度器参数

MOS_ENABLE_SCHEDULER=true

How to Reproduce | 如何重现

直接调用的写接口

Environment | 环境信息

Python

Additional Context | 其他信息

No response

Willingness to Implement | 实现意愿

• I'm willing to implement this myself | 我愿意自己解决
• I would like someone else to implement this | 我希望其他人来解决

[bjmajic]

看了一下有报错：`INFO: 172.20.0.5:35530 - "POST /product/add HTTP/1.1" 200 OK
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory df21fb2d-b721-4855-ab32-d3149b6e3568: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 20e38b80-71cd-4c11-9968-6ab0b60999eb: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 5c974237-e698-4cb8-8fe9-814e4664c132: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory bf762ea0-5a3c-4787-beb3-4ed4cc5270d2: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 86987ebb-e66a-40ea-a2aa-562f61e36b6f: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 35e7a537-c44e-453b-a8ec-4cd987824067: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 563c449d-ff39-456d-a827-f4fa3365e1e6: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory a865a671-9a73-40ba-9a46-8fad22f5ef29: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory c1e2323b-17cf-477a-ad9e-df299c97a42c: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 5d33a551-2dc6-4484-8851-28bc6b566c88: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3ef...a2d-8d83-14a5b9f73dc6"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_scheduler.task_schedule_modules.handlers.mem_read_handler - WARNING - mem_read_handler.py:146 - _process_memories_with_reader - [_process_memories_with_reader] Failed to get memory 983bf703-b051-4f1a-ad74-67fa9ed5dec5: 1 validation error for TreeNodeTextualMemoryMetadata
internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "0f4...9b3-aa27-a6bd76bcb191"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type
31da2f81983126c74786a3fdaa1b8378 | path=/product/add | memos.mem_reader.read_skill_memory.process_skill_memory - WARNING - process_skill_memory.py:978 - _get_skill_file_storage_location - Environment variable [SKILLS_REPO_BACKEND] is invalid, using LOCAL to store skill
8e33e6c375fb30c13680b3bc5df4e5be | path=/product/search | SearchHandler - ERROR - single_cube.py:220 - _search_text - Error in search_text: 1 validation error for TextualMemoryItem
metadata.internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3af...ab6-b0bd-c74ec9f2e22a"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type; traceback: Traceback (most recent call last):
File "/app/src/memos/multi_mem_cube/single_cube.py", line 209, in _search_text
text_memories = self._fast_search(search_req, user_context)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/multi_mem_cube/single_cube.py", line 400, in _fast_search
search_results = search_text_memories(
^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/search/search_service.py", line 79, in search_text_memories
return text_mem.search(
^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree.py", line 214, in search
return searcher.search(
^^^^^^^^^^^^^^^^
File "/app/src/memos/utils.py", line 237, in wrapper
result = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/searcher.py", line 243, in search
retrieved_results = self.retrieve(
^^^^^^^^^^^^^^
File "/app/src/memos/utils.py", line 237, in wrapper
result = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/searcher.py", line 136, in retrieve
results = self._retrieve_paths(
^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/utils.py", line 237, in wrapper
result = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/searcher.py", line 498, in _retrieve_paths
results.extend(t.result())
^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 456, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
raise self._exception
File "/usr/local/lib/python3.11/concurrent/futures/thread.py", line 58, in run
result = self.fn(*self.args, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/context/context.py", line 275, in wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/utils.py", line 237, in wrapper
result = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/searcher.py", line 813, in _retrieve_from_long_term_and_user
results.extend(task.result())
^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 456, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
raise self._exception
File "/usr/local/lib/python3.11/concurrent/futures/thread.py", line 58, in run
result = self.fn(*self.args, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/context/context.py", line 275, in wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/recall.py", line 172, in retrieve
vector_results = future_vector.result()
^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 456, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
raise self._exception
File "/usr/local/lib/python3.11/concurrent/futures/thread.py", line 58, in run
result = self.fn(*self.args, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/context/context.py", line 275, in wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/recall.py", line 514, in _vector_recall
return [TextualMemoryItem.from_dict(n) for n in ordered_nodes]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/tree_text_memory/retrieve/recall.py", line 514, in
return [TextualMemoryItem.from_dict(n) for n in ordered_nodes]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/memos/memories/textual/item.py", line 325, in from_dict
return cls(**data)
^^^^^^^^^^^
File "/usr/local/lib/python3.11/site-packages/pydantic/main.py", line 250, in init
validated_self = self.pydantic_validator.validate_python(data, self_instance=self)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for TextualMemoryItem
metadata.internal_info
Input should be a valid dictionary [type=dict_type, input_value='{"ingest_batch_id": "3af...ab6-b0bd-c74ec9f2e22a"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.12/v/dict_type

[bjmajic]

经检测：写入阶段（错误在此埋下）：POST /product/add 异步写入时，mem_reader（multi_modal_struct.py）会给每个 raw chunk 记忆打上 internal_info = {"ingest_batch_id": ..., "chunk_index": ..., "chunk_total": ...} —— 这是一个 dict。随后落到 Neo4j 时：
def _sanitize_neo4j_value(value: Any) -> Any:
"""Convert values unsupported by Neo4j properties into safe serializations."""
...
if isinstance(value, dict):
return json.dumps(value, ensure_ascii=False, sort_keys=True)
因为 Neo4j 节点属性不支持嵌套 map，internal_info 被 json.dumps 成了 JSON 字符串存进去。

读取阶段（报错爆发）：读回来时 get_node/get_nodes 只对 sources 做了 json.loads 反序列化（neo4j_community.py:1165-1173），没有处理 internal_info，于是它以字符串形式进入 pydantic
internal_info: dict | None = Field(
default=None,
description="Internal algorithm metadata reserved for system use.",
)
dict | None 校验字符串 → ValidationError。这解释了两条报错路径：

mem_read_handler：MOS_ENABLE_SCHEDULER=true 时，异步 add 的后台精炼任务 text_mem.get(mem_id) 逐条取回 raw chunk → 全部失败（所以你看到十几条 WARNING）；
/product/search fast 模式：recall._vector_recall → graph_store.get_nodes → TextualMemoryItem.from_dict(n) → 同样的校验失败。
而 /product/add 本身返回 200，因为写入时 dict 是合法的，只有读路径才会炸。

[bjmajic]

还有一个问题，完成记忆抽取后，Qrant 中的死向量没有被同步删除，会占用召回的 top k 位置：Point dc84f1af-307e-4450-8656-aa3ebcf15f29

Payload:

memory
user: [10:59 AM on 28 August, 2026]: 今天北京天气如何,适合穿什么衣服 assistant: [10:59 AM on 28 August, 2026]: 今天(8月28日,周五)北京天气和穿衣建议如下: 🌧️ 今日天气 - 天气:阴有阵雨转多云,早晨全市出现小雨,路面湿滑;白天仍有阵雨,午后转多云 - 气温:最高约 24℃,夜间最低约 19℃(受降雨影响,明显比昨天凉快) - 风力:北转南风 2~3 级,风力不大 - 空气质量:优 👕 穿衣建议 - 今天偏凉(尤其早上降雨时体感更凉),建议穿 长袖T恤 / 薄衬衫 + 长裤 或单裤 - 怕凉的、老人和小孩建议外加 一件薄外套/针织开衫 - 一定要 带伞,出门注意防滑 - 降雨时段湿度较大,体质弱的朋友留意别着凉 📅 给周末的参考:明后天转晴,最高回升到 31℃ 左右,中午会热起来,紫外线也强,出门注意防晒;早晚温差拉大(约 13℃),早晚凉、中午热,方便的话可以带件薄外套随时增减。 (数据来源:中国天气网,8月28日早间更新)

vector_sync
success

user_id
default

session_id
20260828_105748_8243c4

status
activated

evolve_to
[]

version
1

history
[]

type
fact

key
user: [10:59 AM on 28 August, 2026]: 今天北京天气如何,适合穿什么衣服

confidence
0.99

tags
[
0:"mode:fast"
]

updated_at
2026-08-28T10:59:07.100163

memory_type
LongTermMemory

sources
[
0:"{"type": "chat", "role": "user", "chat_time": "10:…"
1:"{"type": "chat", "role": "assistant", "chat_time":…"
]

created_at
2026-08-28T10:59:06.811568

usage
[]

background
[working_binding:dc84f1af-307e-4450-8656-aa3ebcf15f29] direct built from raw inputs

file_ids
[]

working_binding
dc84f1af-307e-4450-8656-aa3ebcf15f29

is_fast
true

user_name
default

delete_time

delete_record_id

Vectors:
Vectors:
Default vector
Length:
1024

Point fc13096e-6a78-4967-8fed-e799eb4a1e1b

Payload:

memory
2026年8月28日上午10:59，用户查询了北京当天的天气和穿衣建议。助手回复称当天北京阴有阵雨转多云，最高气温约24℃，最低约19℃，风力北转南风2~3级，空气质量优；建议穿长袖T恤、薄衬衫加长裤，怕凉者可加薄外套，并提醒带伞防滑。助手还提供了周末天气参考：明后天转晴，最高约31℃，紫外线强，早晚温差约13℃，建议带薄外套。

vector_sync
success

user_id
default

session_id
20260828_105748_8243c4

status
activated

evolve_to
[]

version
1

history
[]

type
fact

key
北京天气查询与穿衣建议

confidence
0.99

tags
[
0:"天气"
1:"穿衣建议"
2:"北京"
3:"2026年8月"
]

updated_at
2026-08-28T10:59:22.309989

memory_type
LongTermMemory

sources
[
0:"{"type": "chat", "role": "user", "chat_time": "10:…"
1:"{"type": "chat", "role": "assistant", "chat_time":…"
]

created_at
2026-08-28T10:59:22.307865

usage
[]

background
2026年8月28日上午，用户查询了北京当天的天气和穿衣建议。助手告知当天阴有阵雨转多云，气温约19-24℃，建议穿长袖加长裤，并带伞；同时提到周末转晴升温至31℃左右，早晚温差大，需注意防晒和增减衣物。用户可能计划在当天外出或安排活动，因此关注天气细节。

file_ids
[]

working_binding
fc13096e-6a78-4967-8fed-e799eb4a1e1b

user_name
default

delete_time

delete_record_id

Vectors:
Vectors:
Default vector
Length:
1024