MemoryOS 2.0.19__tar.gz → 2.0.22__tar.gz

{memoryos-2.0.19 → memoryos-2.0.22}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: MemoryOS
-Version: 2.0.19
+Version: 2.0.22
 Summary: Intelligence Begins with Memory
 License: Apache-2.0
 License-File: LICENSE
@@ -103,10 +103,10 @@ Description-Content-Type: text/markdown
 </p>
 
 <p align="center">
-  <strong>🎯 +43.70% Accuracy vs. OpenAI Memory</strong><br/>
-  <strong>🏆 Top-tier Long-term Memory + Personalization</strong><br/>
-  <strong>💰 Saves 35.24% Memory Tokens</strong><br/>
-  <sub>LoCoMo 75.80 • LongMemEval +40.43% • PrefEval-10 +2568% • PersonaMem +40.75%</sub>
+  <strong>🏆 Leading Performance Across Agent and User Memory Benchmarks</strong><br/>
+  <strong>🤖 OpenClaw Task Completion Improves from 36.63% to 50.87%</strong><br/>
+  <strong>🎯 92.34 on LoCoMo and 93.40 on LongMemEval</strong><br/>
+  <strong>📊 Unified Evaluation Across 14 Commercial Memory Products</strong>
   <!-- <a href="https://memos.openmem.net/">
     <img src="https://statics.memtensor.com.cn/memos/github_api_free_banner.gif" alt="MemOS Free API Banner">
   </a> -->
@@ -155,6 +155,9 @@ Your lobsters and Hermes Agents now have **the best** memory system — choose *
 
 ### News
 
+- **2026-07-02** · 🏆 **MemOS Advances Agent and User Memory Benchmarks**
+  With MemOS, **OpenClaw** improves average task completion from **36.63% to 50.87%** across five agent tasks. MemOS also achieves **92.34 on LoCoMo** and **93.40 on LongMemEval**, and leads in **OmniMemEval**, a unified evaluation of 14 commercial memory products across ten datasets.
+
 - **2026-05-09** · 🧠 **memos-local-plugin 2.0**
   Official local memory plugin for **Hermes Agent** and **OpenClaw**. One core powers self-evolving memory across L1 traces, L2 policies, L3 world models, and crystallized Skills, with local-first storage and feedback-driven retrieval.
 
@@ -357,14 +360,8 @@ MemOS is a **Memory Operating System** for LLMs and AI agents that unifies store
 
 ### What are the benchmark results?
 
-| Benchmark | MemOS Result | Improvement |
-|-----------|--------------|-------------|
-| LoCoMo | 75.80 | - |
-| LongMemEval | +40.43% vs baseline | - |
-| PrefEval-10 | +2568% | - |
-| PersonaMem | +40.75% | - |
-| **vs OpenAI Memory** | +43.70% Accuracy | - |
-| **Token Savings** | 35.24% | - |
+MemOS achieves **92.34 on LoCoMo** and **93.40 on LongMemEval** for User Memory, while improving **OpenClaw** average task completion from **36.63% to 50.87%** across five Agent Memory tasks. For details, see [OmniMemEval](https://github.com/MemTensor/OmniMemEval), our unified evaluation framework for benchmarking 14 commercial memory products across ten datasets.
+
 
 ### How does MemOS compare to other memory solutions?
 

{memoryos-2.0.19 → memoryos-2.0.22}/README.md

@@ -19,10 +19,10 @@
 </p>
 
 <p align="center">
-  <strong>🎯 +43.70% Accuracy vs. OpenAI Memory</strong><br/>
-  <strong>🏆 Top-tier Long-term Memory + Personalization</strong><br/>
-  <strong>💰 Saves 35.24% Memory Tokens</strong><br/>
-  <sub>LoCoMo 75.80 • LongMemEval +40.43% • PrefEval-10 +2568% • PersonaMem +40.75%</sub>
+  <strong>🏆 Leading Performance Across Agent and User Memory Benchmarks</strong><br/>
+  <strong>🤖 OpenClaw Task Completion Improves from 36.63% to 50.87%</strong><br/>
+  <strong>🎯 92.34 on LoCoMo and 93.40 on LongMemEval</strong><br/>
+  <strong>📊 Unified Evaluation Across 14 Commercial Memory Products</strong>
   <!-- <a href="https://memos.openmem.net/">
     <img src="https://statics.memtensor.com.cn/memos/github_api_free_banner.gif" alt="MemOS Free API Banner">
   </a> -->
@@ -71,6 +71,9 @@ Your lobsters and Hermes Agents now have **the best** memory system — choose *
 
 ### News
 
+- **2026-07-02** · 🏆 **MemOS Advances Agent and User Memory Benchmarks**
+  With MemOS, **OpenClaw** improves average task completion from **36.63% to 50.87%** across five agent tasks. MemOS also achieves **92.34 on LoCoMo** and **93.40 on LongMemEval**, and leads in **OmniMemEval**, a unified evaluation of 14 commercial memory products across ten datasets.
+
 - **2026-05-09** · 🧠 **memos-local-plugin 2.0**
   Official local memory plugin for **Hermes Agent** and **OpenClaw**. One core powers self-evolving memory across L1 traces, L2 policies, L3 world models, and crystallized Skills, with local-first storage and feedback-driven retrieval.
 
@@ -273,14 +276,8 @@ MemOS is a **Memory Operating System** for LLMs and AI agents that unifies store
 
 ### What are the benchmark results?
 
-| Benchmark | MemOS Result | Improvement |
-|-----------|--------------|-------------|
-| LoCoMo | 75.80 | - |
-| LongMemEval | +40.43% vs baseline | - |
-| PrefEval-10 | +2568% | - |
-| PersonaMem | +40.75% | - |
-| **vs OpenAI Memory** | +43.70% Accuracy | - |
-| **Token Savings** | 35.24% | - |
+MemOS achieves **92.34 on LoCoMo** and **93.40 on LongMemEval** for User Memory, while improving **OpenClaw** average task completion from **36.63% to 50.87%** across five Agent Memory tasks. For details, see [OmniMemEval](https://github.com/MemTensor/OmniMemEval), our unified evaluation framework for benchmarking 14 commercial memory products across ten datasets.
+
 
 ### How does MemOS compare to other memory solutions?
 

{memoryos-2.0.19 → memoryos-2.0.22}/pyproject.toml

@@ -4,7 +4,7 @@
 ##############################################################################
 
 name = "MemoryOS"
-version = "2.0.19"
+version = "2.0.22"
 description = "Intelligence Begins with Memory"
 license = {text = "Apache-2.0"}
 readme = "README.md"
@@ -89,6 +89,8 @@ mem-user = [
 
 # MemReader
 mem-reader = [
+    # chonkie>=1.6 pulls numpy>=2.0; any environment that also has scipy
+    # installed must use scipy>=1.13 (older scipy pins numpy<1.27). See #1342.
     "chonkie (>=1.0.7,<2.0.0)",  # Sentence chunking library
     "markitdown[docx,pdf,pptx,xls,xlsx] (>=0.1.1,<0.2.0)",  # Markdown parser for various file formats
     "langchain-text-splitters (>=1.0.0,<2.0.0)", # markdown chunk for langchain
@@ -193,7 +195,9 @@ zep-cloud = "^2.15.0"
 rouge-score = "^0.1.2"
 nltk = "^3.9.1"
 bert-score = "^0.3.13"
-scipy = "^1.10.1"
+# scipy 1.13+ supports numpy 2.x. Older scipy (1.10/1.11/1.12) pins numpy<1.27
+# which collides with chonkie>=1.6 (requires numpy>=2.0.0) — see issue #1342.
+scipy = "^1.13.0"
 python-dotenv = "^1.1.1"
 langgraph = "^0.5.1"
 

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.0.19"
+__version__ = "2.0.22"
 
 from memos.configs.mem_cube import GeneralMemCubeConfig
 from memos.configs.mem_os import MOSConfig

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/config.py

@@ -979,6 +979,10 @@ class APIConfig:
                     "general_llm": APIConfig.get_memreader_general_llm_config(),
                     # Image parser LLM (requires vision model)
                     "image_parser_llm": APIConfig.get_image_parser_llm_config(),
+                    # Preference extractor LLM. Reader falls back to general_llm when unset.
+                    "preference_extractor_llm": APIConfig.get_preference_extractor_llm_config()
+                    if os.getenv("PREFERENCE_EXTRACTOR_MODEL")
+                    else None,
                     "embedder": APIConfig.get_embedder_config(),
                     "chunker": {
                         "backend": "sentence",
@@ -1108,6 +1112,10 @@ class APIConfig:
                     "general_llm": APIConfig.get_memreader_general_llm_config(),
                     # Image parser LLM (requires vision model)
                     "image_parser_llm": APIConfig.get_image_parser_llm_config(),
+                    # Preference extractor LLM. Reader falls back to general_llm when unset.
+                    "preference_extractor_llm": APIConfig.get_preference_extractor_llm_config()
+                    if os.getenv("PREFERENCE_EXTRACTOR_MODEL")
+                    else None,
                     "embedder": APIConfig.get_embedder_config(),
                     "chunker": {
                         "backend": "sentence",

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/handlers/chat_handler.py

@@ -1225,7 +1225,13 @@ class ChatHandler(BaseHandler):
                 [f"{msg['role']}: {msg['content']}" for msg in current_messages[-2:]]
             )
             further_suggestion_prompt = FURTHER_SUGGESTION_PROMPT.format(dialogue=dialogue_info)
-            message_list = [{"role": "system", "content": further_suggestion_prompt}]
+            message_list = [
+                {
+                    "role": "system",
+                    "content": "You are a helpful assistant that generates suggestion queries based on dialogue context.",
+                },
+                {"role": "user", "content": further_suggestion_prompt},
+            ]
             response = self.llm.generate(message_list)
             clean_response = clean_json_response(response)
             response_json = json.loads(clean_response)

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/handlers/formatters_handler.py

@@ -13,6 +13,13 @@ from memos.templates.instruction_completion import instruct_completion
 
 logger = get_logger(__name__)
 
+__all__ = [
+    "format_memory_item",
+    "post_process_textual_mem",
+    "rerank_knowledge_mem",
+    "to_iter",
+]
+
 
 def to_iter(running: Any) -> list[Any]:
     """
@@ -164,6 +171,7 @@ def rerank_knowledge_mem(
     text_mem: list[dict[str, Any]],
     top_k: int,
     file_mem_proportion: float = 0.5,
+    strip_conversation_sources: bool = False,
 ) -> list[dict[str, Any]]:
     """
     Rerank knowledge memories and keep conversation memories.
@@ -193,8 +201,9 @@ def rerank_knowledge_mem(
         item["memory"] = item["metadata"]["sources"][0]["content"]
         item["metadata"]["sources"] = []
 
-    for item in conversation_mem:
-        item.setdefault("metadata", {})["sources"] = []
+    if strip_conversation_sources:
+        for item in conversation_mem:
+            item.setdefault("metadata", {})["sources"] = []
 
     # deduplicate: remove items with duplicate memory content
     original_count = len(reranked_knowledge_mem)

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/handlers/scheduler_handler.py

@@ -146,18 +146,50 @@ def handle_scheduler_allstatus(
         sched_failed = all_tasks_summary.failed
         sched_cancelled = all_tasks_summary.cancelled
 
-        # If queue monitor is available, prefer its live waiting/in_progress counts
+        # If queue monitor is available, prefer its live waiting/in_progress counts.
+        #
+        # Two queue backends produce different ``queue_status_data`` shapes:
+        #   - Redis queue: ``{"running": .., "remaining": .., "pending": ..,
+        #     "<stream_key_prefix>:<user>:<cube>:<label>": {"running": ..,
+        #     "remaining": .., "pending": ..}, ...}`` — the per-stream entries
+        #     are nested dicts and their keys start with the scheduler stream
+        #     prefix (default ``"scheduler:messages:stream:v2.0:"``). The loop
+        #     below aggregates those per-stream entries.
+        #   - Local in-memory queue: only the top-level totals
+        #     ``{"running", "remaining", "pending"}`` are present (see
+        #     ``TaskScheduleMonitor._get_local_tasks_status``). There are no
+        #     per-stream keys to enumerate, so the prefix loop would silently
+        #     collapse the summary to all zeros. Detect that case and fall back
+        #     to the flat totals so Docker / no-Redis deployments still surface
+        #     accurate in_progress / waiting / pending counts (issue #1395).
         if mem_scheduler.task_schedule_monitor:
             queue_status_data = mem_scheduler.task_schedule_monitor.get_tasks_status() or {}
             scheduler_waiting = 0
             scheduler_in_progress = 0
             scheduler_pending = 0
-            for key, value in queue_status_data.items():
-                if not key.startswith("scheduler:"):
-                    continue
-                scheduler_in_progress += int(value.get("running", 0) or 0)
-                scheduler_pending += int(value.get("pending", value.get("remaining", 0)) or 0)
-                scheduler_waiting += int(value.get("remaining", 0) or 0)
+            per_stream_keys = [
+                key
+                for key, value in queue_status_data.items()
+                if isinstance(key, str) and key.startswith("scheduler:") and isinstance(value, dict)
+            ]
+
+            if per_stream_keys:
+                for key in per_stream_keys:
+                    value = queue_status_data[key]
+                    scheduler_in_progress += int(value.get("running", 0) or 0)
+                    scheduler_pending += int(value.get("pending", value.get("remaining", 0)) or 0)
+                    scheduler_waiting += int(value.get("remaining", 0) or 0)
+            elif not getattr(mem_scheduler, "use_redis_queue", True):
+                # Local-queue mode: there are no per-stream keys; use the flat
+                # totals from ``_get_local_tasks_status``. ``remaining`` reflects
+                # queued depth (waiting / pending) and ``running`` reflects the
+                # dispatcher's in-flight count.
+                scheduler_in_progress = int(queue_status_data.get("running", 0) or 0)
+                scheduler_waiting = int(queue_status_data.get("remaining", 0) or 0)
+                scheduler_pending = int(
+                    queue_status_data.get("pending", queue_status_data.get("remaining", 0)) or 0
+                )
+
             sched_waiting = scheduler_waiting
             sched_in_progress = scheduler_in_progress
             sched_pending = scheduler_pending

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/handlers/search_handler.py

@@ -118,6 +118,23 @@ class SearchHandler(BaseHandler):
             top_k=search_req_local.top_k,
             file_mem_proportion=0.5,
         )
+        hooked_results = trigger_hook(
+            H.SEARCH_RESULTS_AFTER_RERANK,
+            handler=self,
+            search_req=search_req_local,
+            results=results,
+        )
+        if hooked_results is not None:
+            results = hooked_results
+
+        hooked_results = trigger_hook(
+            H.SEARCH_CONTEXT_RENDER,
+            handler=self,
+            search_req=search_req_local,
+            results=results,
+        )
+        if hooked_results is not None:
+            results = hooked_results
 
         self.logger.info(
             f"[SearchHandler] Final search results: count={len(results)} results={results}"

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/handlers/suggestion_handler.py

@@ -49,7 +49,13 @@ def _get_further_suggestion(
                 ]
             )
         further_suggestion_prompt = FURTHER_SUGGESTION_PROMPT.format(dialogue=dialogue_info)
-        message_list = [{"role": "system", "content": further_suggestion_prompt}]
+        message_list = [
+            {
+                "role": "system",
+                "content": "You are a helpful assistant that generates suggestion queries based on dialogue context.",
+            },
+            {"role": "user", "content": further_suggestion_prompt},
+        ]
         response = llm.generate(message_list)
         clean_response = clean_json_response(response)
         response_json = json.loads(clean_response)
@@ -111,7 +117,13 @@ def handle_get_suggestion_queries(
             memories = "\n".join([m.memory[:200] for m in text_mem_results])
 
         # Generate suggestions using LLM
-        message_list = [{"role": "system", "content": suggestion_prompt.format(memories=memories)}]
+        message_list = [
+            {
+                "role": "system",
+                "content": "You are a helpful assistant that generates suggestion queries based on the user's recent memories.",
+            },
+            {"role": "user", "content": suggestion_prompt.format(memories=memories)},
+        ]
         response = llm.generate(message_list)
         clean_response = clean_json_response(response)
         response_json = json.loads(clean_response)

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/api/product_models.py

@@ -422,6 +422,14 @@ class APISearchRequest(BaseRequest):
         ),
     )
 
+    rerank: bool = Field(
+        True,
+        description=(
+            "Whether to apply the textual memory reranker during search. "
+            "Set false to return retrieval-order candidates before post-search dedup/formatting."
+        ),
+    )
+
     pref_top_k: int = Field(
         6,
         ge=0,
@@ -464,6 +472,11 @@ class APISearchRequest(BaseRequest):
         description="Number of skill memories to retrieve (top-K). Default: 3.",
     )
 
+    context_format: str = Field(
+        "memory",
+        description="Optional search context format passed through to installed plugins.",
+    )
+
     # ==== Filter conditions ====
     # TODO: maybe add detailed description later
     filter: dict[str, Any] | None = Field(

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/configs/mem_reader.py

@@ -36,6 +36,10 @@ class BaseMemReaderConfig(BaseConfig):
         default=None,
         description="Vision LLM for image parsing. Falls back to general_llm if not set.",
     )
+    preference_extractor_llm: LLMConfigFactory | None = Field(
+        default=None,
+        description="LLM for preference extraction. Falls back to general_llm if not set.",
+    )
     embedder: EmbedderConfigFactory = Field(
         ..., description="Embedder configuration for the MemReader"
     )

memoryos-2.0.22/src/memos/dream/README.md

@@ -0,0 +1,129 @@
+# Dream Plugin
+
+Dream is an optional MemOS community feature, currently in beta. It explores how
+an agent can reflect on recently added memories outside the foreground request
+path, then consolidate them into higher-level context, insights, and diary
+entries. The implementation is intentionally lightweight and extensible, and we
+welcome the community to help improve the signal policy, prompts, persistence,
+diary experience, and downstream integrations.
+
+## Status and Enablement
+
+Dream is disabled by default.
+
+With no plugin environment variables configured, MemOS loads ordinary plugins as
+before, but it does not load the built-in `dream` plugin. Enable it explicitly:
+
+```bash
+MEMOS_ENABLED_PLUGINS=dream
+```
+
+`MEMOS_DISABLED_PLUGINS` has the highest priority, so this keeps Dream disabled:
+
+```bash
+MEMOS_ENABLED_PLUGINS=dream
+MEMOS_DISABLED_PLUGINS=dream
+```
+
+## Current Capabilities
+
+When enabled, the built-in `CommunityDreamPlugin` provides:
+
+- signal capture from successful memory add operations;
+- deterministic Dream metadata enrichment during fine extraction;
+- scheduler-driven Dream execution through the `dream.execute` hook;
+- manual cube-level triggering through `POST /dream/trigger/cube`;
+- Dream diary querying through `POST /dream/diary`;
+- `Context` recall merged into normal search results.
+
+The default pipeline is:
+
+1. Build or update `Context` nodes from pending memories.
+2. Form Dream motives from recently added source memories.
+3. Recall related `UserMemory` and `LongTermMemory` nodes.
+4. Use the configured LLM to produce at most one insight per motive.
+5. Generate a human-readable Dream diary entry.
+6. Persist valid insight actions and diary entries to the graph database.
+
+If no LLM is available, Dream can still run the pipeline, but fallback reasoning
+does not write new insight memories because zero-confidence actions are skipped.
+
+## Usage
+
+Enable Dream and start the API service:
+
+```bash
+export MEMOS_ENABLED_PLUGINS=dream
+make serve
+```
+
+Check plugin health:
+
+```bash
+curl http://127.0.0.1:8000/dream/diary/health
+```
+
+Manually submit a cube-level Dream task:
+
+```bash
+curl -X POST "http://127.0.0.1:8000/dream/trigger/cube?cube_id=<cube_id>&user_id=<user_id>&user_name=<user_name>"
+```
+
+Query recent Dream diary entries:
+
+```bash
+curl -X POST http://127.0.0.1:8000/dream/diary \
+  -H "Content-Type: application/json" \
+  -d '{"cube_id": "<cube_id>", "filter": {"limit": 5}}'
+```
+
+Fetch one diary entry:
+
+```bash
+curl -X POST http://127.0.0.1:8000/dream/diary \
+  -H "Content-Type: application/json" \
+  -d '{"cube_id": "<cube_id>", "filter": {"task_id": "dream_diary_xxx"}}'
+```
+
+## Configuration
+
+Plugin loading:
+
+- `MEMOS_ENABLED_PLUGINS=dream`: enable Dream.
+- `MEMOS_DISABLED_PLUGINS=dream`: disable Dream, even if it is also enabled.
+
+Dream-specific options:
+
+- `MEMOS_DREAM_HEURISTIC_ENRICHER`: defaults to `on`.
+- `MEMOS_DREAM_ENRICH_OVERWRITE`: defaults to `off`.
+- `MEMOS_DREAM_CONTEXT_ENABLED`: defaults to `on`.
+- `MEMOS_DREAM_CONTEXT_SUMMARY_LLM`: defaults to `on`.
+- `MEMOS_DREAM_CONTEXT_BINDING_LLM`: defaults to `on`.
+- `MEMOS_DREAM_CONTEXT_BINDING_MIN_GROUP_SIZE`: defaults to `2`.
+- `MEMOS_DREAM_CONTEXT_BINDING_MAX_GROUP_SIZE`: defaults to `30`.
+- `MEMOS_DREAM_CONTEXT_BINDING_CONFIDENCE_THRESHOLD`: defaults to `0.65`.
+
+## Beta Limitations
+
+- Signals are stored in memory and do not survive process restarts.
+- The automatic trigger policy is currently a simple pending-memory threshold.
+- The built-in signal source focuses on new-memory accumulation. Conflict,
+  feedback, frequency, and fragmentation signals are extension points.
+- The diary and surfacing experience is still early.
+- Write-back policies for update, merge, archive, and long-term maintenance are
+  available as extension directions, not complete product behavior.
+
+## Contributing
+
+Dream is designed as a community-building surface. Good places to contribute:
+
+- better trigger policies and signal stores;
+- stronger motive formation and recall strategies;
+- safer and more useful reasoning prompts;
+- richer diary generation and user-facing surfacing;
+- memory lifecycle and maintenance policies;
+- alternative `dream` plugin implementations with higher priority.
+
+Projects can ship their own plugin with the same logical name (`dream`). When
+multiple providers expose `dream`, the plugin manager keeps the implementation
+with the highest priority.

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/dream/plugin.py

@@ -46,6 +46,7 @@ class CommunityDreamPlugin(MemOSPlugin):
     version = "0.1.0"
     description = "Built-in Dream plugin"
     priority = 10
+    enabled_by_default = False
 
     def on_load(self) -> None:
         self.context: dict[str, Any] = {"shared": {}, "configs": {}}

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/mem_os/core.py

@@ -1170,7 +1170,7 @@ class MOSCore:
             bool: True if successful, False otherwise.
         """
         # Validate current user has access to this cube
-        self._validate_cube_access(cube_id, target_user_id)
+        self._validate_cube_access(self.user_id, cube_id)
 
         # Validate target user exists
         if not self.user_manager.validate_user(target_user_id):

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/mem_reader/multi_modal_struct.py

@@ -1035,7 +1035,7 @@ class MultiModalStructMemReader(SimpleStructMemReader):
                     process_preference_fine,
                     non_file_url_fast_items,
                     info,
-                    self.general_llm,
+                    getattr(self, "preference_extractor_llm", self.general_llm),
                     self.embedder,
                     **kwargs,
                 )
@@ -1127,7 +1127,7 @@ class MultiModalStructMemReader(SimpleStructMemReader):
                 process_preference_fine,
                 non_file_url_nodes,
                 info,
-                self.general_llm,
+                getattr(self, "preference_extractor_llm", self.general_llm),
                 self.embedder,
                 **kwargs,
             )

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/mem_reader/read_pref_memory/process_preference_memory.py

@@ -81,6 +81,19 @@ def _extract_implicit_preference(qa_pair_str: str, llm) -> list[dict[str, Any]]
         return None
 
 
+def _normalize_preference_data(preference_data: dict[str, Any]) -> dict[str, Any] | None:
+    preference = str(preference_data.get("preference") or "").strip()
+    context_summary = str(preference_data.get("context_summary") or "").strip()
+
+    if not preference or not context_summary:
+        return None
+
+    normalized_data = preference_data.copy()
+    normalized_data["preference"] = preference
+    normalized_data["context_summary"] = context_summary
+    return normalized_data
+
+
 def _create_preference_memory_item(
     preference_data: dict[str, Any],
     preference_type: str,
@@ -167,8 +180,12 @@ def _process_single_chunk_explicit(
 
     memories = []
     for pref in explicit_pref:
+        normalized_pref = _normalize_preference_data(pref)
+        if not normalized_pref:
+            continue
+
         memory = _create_preference_memory_item(
-            preference_data=pref,
+            preference_data=normalized_pref,
             preference_type="explicit_preference",
             fast_item=fast_item,
             info=info,
@@ -198,8 +215,12 @@ def _process_single_chunk_implicit(
 
     memories = []
     for pref in implicit_pref:
+        normalized_pref = _normalize_preference_data(pref)
+        if not normalized_pref:
+            continue
+
         memory = _create_preference_memory_item(
-            preference_data=pref,
+            preference_data=normalized_pref,
             preference_type="implicit_preference",
             fast_item=fast_item,
             info=info,

{memoryos-2.0.19 → memoryos-2.0.22}/src/memos/mem_reader/simple_struct.py

@@ -184,6 +184,12 @@ class SimpleStructMemReader(BaseMemReader, ABC):
             if config.general_llm is not None
             else self.llm
         )
+        preference_extractor_llm_config = getattr(config, "preference_extractor_llm", None)
+        self.preference_extractor_llm = (
+            LLMFactory.from_config(preference_extractor_llm_config)
+            if preference_extractor_llm_config is not None
+            else self.general_llm
+        )
         self.qwen_llm = None
         qwen_llm_config = getattr(config, "qwen_llm", None)
         if qwen_llm_config:
@@ -286,8 +292,14 @@ class SimpleStructMemReader(BaseMemReader, ABC):
         response_json = self._safe_parse(response_text)
 
         if not response_json:
+            # NOTE: the key MUST be ``"memory list"`` (with a space) — the
+            # downstream consumers in ``_process_chat_data`` /
+            # ``_process_transfer_chat_data`` read via
+            # ``resp.get("memory list", [])``. A typo here drops the
+            # salvaged item silently and causes ``/product/add`` to return
+            # 200 with zero memories written to Neo4j (bug #1355).
             return {
-                "memory_list": [
+                "memory list": [
                     {
                         "key": mem_str[:10],
                         "memory_type": "UserMemory",