Dev 20260224 v2.0.7 (MemTensor#1127)

## Description

Please include a summary of the change, the problem it solves, the
implementation approach, and relevant context. List any dependencies
required for this change.

Related Issue (Required):  Fixes @issue_number

## Type of change

Please delete options that are not relevant.

- [ ] Bug fix (non-breaking change which fixes an issue)
- [ ] New feature (non-breaking change which adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to not work as expected)
- [ ] Refactor (does not change functionality, e.g. code style
improvements, linting)
- [ ] Documentation update

## How Has This Been Tested?

Please describe the tests that you ran to verify your changes. Provide
instructions so we can reproduce. Please also list any relevant details
for your test configuration

- [ ] Unit Test
- [ ] Test Script Or Test Steps (please provide)
- [ ] Pipeline Automated API Test (please provide)

## Checklist

- [ ] I have performed a self-review of my own code | 我已自行检查了自己的代码
- [ ] I have commented my code in hard-to-understand areas |
我已在难以理解的地方对代码进行了注释
- [ ] I have added tests that prove my fix is effective or that my
feature works | 我已添加测试以证明我的修复有效或功能正常
- [ ] I have created related documentation issue/PR in
[MemOS-Docs](https://github.com/MemTensor/MemOS-Docs) (if applicable) |
我已在 [MemOS-Docs](https://github.com/MemTensor/MemOS-Docs) 中创建了相关的文档
issue/PR（如果适用）
- [ ] I have linked the issue to this PR (if applicable) | 我已将 issue
链接到此 PR（如果适用）
- [ ] I have mentioned the person who will review this PR | 我已提及将审查此 PR
的人

## Reviewer Checklist
- [ ] closes #xxxx (Replace xxxx with the GitHub issue number)
- [ ] Made sure Checks passed
- [ ] Tests have been provided

Author: bittergreen

README.md

@@ -345,10 +345,10 @@ url = {https://global-sci.com/article/91443/memory3-language-modeling-with-expli
 
 ## 🙌 Contributing
 
-We welcome contributions from the community! Please read our [contribution guidelines](https://memos-docs.openmem.net/contribution/overview) to get started.
+We welcome contributions from the community! Please read our [contribution guidelines](https://memos-docs.openmem.net/open_source/contribution/overview/) to get started.
 
 <br>
 
 ## 📄 License
 
-MemOS is licensed under the [Apache 2.0 License](./LICENSE).
+MemOS is licensed under the [Apache 2.0 License](./LICENSE).

pyproject.toml

@@ -4,7 +4,7 @@
 ##############################################################################
 
 name = "MemoryOS"
-version = "2.0.6"
+version = "2.0.7"
 description = "Intelligence Begins with Memory"
 license = {text = "Apache-2.0"}
 readme = "README.md"

src/memos/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.0.6"
+__version__ = "2.0.7"
 
 from memos.configs.mem_cube import GeneralMemCubeConfig
 from memos.configs.mem_os import MOSConfig

src/memos/api/handlers/search_handler.py

@@ -64,7 +64,7 @@ def handle_search_memories(self, search_req: APISearchRequest) -> SearchResponse
 
         # Expand top_k for deduplication (5x to ensure enough candidates)
         if search_req_local.dedup in ("sim", "mmr"):
-            search_req_local.top_k = search_req_local.top_k * 5
+            search_req_local.top_k = search_req_local.top_k * 3
 
         # Search and deduplicate
         cube_view = self._build_cube_view(search_req_local)
@@ -152,9 +152,6 @@ def _dedup_text_memories(self, results: dict[str, Any], target_top_k: int) -> di
             return results
 
         embeddings = self._extract_embeddings([mem for _, mem, _ in flat])
-        if embeddings is None:
-            documents = [mem.get("memory", "") for _, mem, _ in flat]
-            embeddings = self.searcher.embedder.embed(documents)
 
         similarity_matrix = cosine_similarity_matrix(embeddings)
 
@@ -235,12 +232,39 @@ def _mmr_dedup_text_memories(
         if len(flat) <= 1:
             return results
 
+        total_by_type: dict[str, int] = {"text": 0, "preference": 0}
+        existing_by_type: dict[str, int] = {"text": 0, "preference": 0}
+        missing_by_type: dict[str, int] = {"text": 0, "preference": 0}
+        missing_indices: list[int] = []
+        for idx, (mem_type, _, mem, _) in enumerate(flat):
+            if mem_type not in total_by_type:
+                total_by_type[mem_type] = 0
+                existing_by_type[mem_type] = 0
+                missing_by_type[mem_type] = 0
+            total_by_type[mem_type] += 1
+
+            embedding = mem.get("metadata", {}).get("embedding")
+            if embedding:
+                existing_by_type[mem_type] += 1
+            else:
+                missing_by_type[mem_type] += 1
+                missing_indices.append(idx)
+
+        self.logger.info(
+            "[SearchHandler] MMR embedding metadata scan: total=%s total_by_type=%s existing_by_type=%s missing_by_type=%s",
+            len(flat),
+            total_by_type,
+            existing_by_type,
+            missing_by_type,
+        )
+        if missing_indices:
+            self.logger.warning(
+                "[SearchHandler] MMR embedding metadata missing; will compute missing embeddings: missing_total=%s",
+                len(missing_indices),
+            )
+
         # Get or compute embeddings
         embeddings = self._extract_embeddings([mem for _, _, mem, _ in flat])
-        if embeddings is None:
-            self.logger.warning("[SearchHandler] Embedding is missing; recomputing embeddings")
-            documents = [mem.get("memory", "") for _, _, mem, _ in flat]
-            embeddings = self.searcher.embedder.embed(documents)
 
         # Compute similarity matrix using NumPy-optimized method
         # Returns numpy array but compatible with list[i][j] indexing
@@ -404,14 +428,32 @@ def _max_similarity(
             return 0.0
         return max(similarity_matrix[index][j] for j in selected_indices)
 
-    @staticmethod
-    def _extract_embeddings(memories: list[dict[str, Any]]) -> list[list[float]] | None:
+    def _extract_embeddings(self, memories: list[dict[str, Any]]) -> list[list[float]]:
         embeddings: list[list[float]] = []
-        for mem in memories:
-            embedding = mem.get("metadata", {}).get("embedding")
-            if not embedding:
-                return None
-            embeddings.append(embedding)
+        missing_indices: list[int] = []
+        missing_documents: list[str] = []
+
+        for idx, mem in enumerate(memories):
+            metadata = mem.get("metadata")
+            if not isinstance(metadata, dict):
+                metadata = {}
+                mem["metadata"] = metadata
+
+            embedding = metadata.get("embedding")
+            if embedding:
+                embeddings.append(embedding)
+                continue
+
+            embeddings.append([])
+            missing_indices.append(idx)
+            missing_documents.append(mem.get("memory", ""))
+
+        if missing_indices:
+            computed = self.searcher.embedder.embed(missing_documents)
+            for idx, embedding in zip(missing_indices, computed, strict=False):
+                embeddings[idx] = embedding
+                memories[idx]["metadata"]["embedding"] = embedding
+
         return embeddings
 
     @staticmethod

src/memos/api/middleware/__init__.py

@@ -1,13 +1,14 @@
 """Krolik middleware extensions for MemOS."""
 
-from .auth import verify_api_key, require_scope, require_admin, require_read, require_write
+from .auth import require_admin, require_read, require_scope, require_write, verify_api_key
 from .rate_limit import RateLimitMiddleware
 
+
 __all__ = [
-    "verify_api_key",
-    "require_scope",
+    "RateLimitMiddleware",
     "require_admin",
     "require_read",
+    "require_scope",
     "require_write",
-    "RateLimitMiddleware",
+    "verify_api_key",
 ]

src/memos/api/product_models.py

@@ -99,12 +99,12 @@ class ChatRequest(BaseRequest):
     manager_user_id: str | None = Field(None, description="Manager User ID")
     project_id: str | None = Field(None, description="Project ID")
     relativity: float = Field(
-        0.0,
+        0.45,
         ge=0,
         description=(
             "Relevance threshold for recalled memories. "
             "Only memories with metadata.relativity >= relativity will be returned. "
-            "Use 0 to disable threshold filtering. Default: 0.3."
+            "Use 0 to disable threshold filtering. Default: 0.45."
         ),
     )
 
@@ -339,12 +339,12 @@ class APISearchRequest(BaseRequest):
     )
 
     relativity: float = Field(
-        0.0,
+        0.45,
         ge=0,
         description=(
             "Relevance threshold for recalled memories. "
             "Only memories with metadata.relativity >= relativity will be returned. "
-            "Use 0 to disable threshold filtering. Default: 0.3."
+            "Use 0 to disable threshold filtering. Default: 0.45."
         ),
     )
 
@@ -785,12 +785,12 @@ class APIChatCompleteRequest(BaseRequest):
     manager_user_id: str | None = Field(None, description="Manager User ID")
     project_id: str | None = Field(None, description="Project ID")
     relativity: float = Field(
-        0.0,
+        0.45,
         ge=0,
         description=(
             "Relevance threshold for recalled memories. "
             "Only memories with metadata.relativity >= relativity will be returned. "
-            "Use 0 to disable threshold filtering. Default: 0.3."
+            "Use 0 to disable threshold filtering. Default: 0.45."
         ),
     )
 

src/memos/api/utils/api_keys.py

@@ -5,8 +5,8 @@
 """
 
 import hashlib
-import os
 import secrets
+
 from dataclasses import dataclass
 from datetime import datetime, timedelta
 

src/memos/embedders/universal_api.py

@@ -73,7 +73,6 @@ async def _create_embeddings():
                     )
                 )
                 logger.info(f"Embeddings request succeeded with {time.time() - init_time} seconds")
-                logger.info(f"Embeddings request response: {response}")
                 return [r.embedding for r in response.data]
             except Exception as e:
                 if self.use_backup_client:

src/memos/graph_dbs/base.py

@@ -1,12 +1,35 @@
+import re
+
 from abc import ABC, abstractmethod
 from typing import Any, Literal
 
 
+# Pattern for valid field names: alphanumeric and underscores, must start with letter or underscore
+_VALID_FIELD_NAME_RE = re.compile(r"^[a-zA-Z_][a-zA-Z0-9_]*$")
+
+
 class BaseGraphDB(ABC):
     """
     Abstract base class for a graph database interface used in a memory-augmented RAG system.
     """
 
+    @staticmethod
+    def _validate_return_fields(return_fields: list[str] | None) -> list[str]:
+        """Validate and sanitize return_fields to prevent query injection.
+
+        Only allows alphanumeric characters and underscores in field names.
+        Silently drops invalid field names.
+
+        Args:
+            return_fields: List of field names to validate.
+
+        Returns:
+            List of valid field names.
+        """
+        if not return_fields:
+            return []
+        return [f for f in return_fields if _VALID_FIELD_NAME_RE.match(f)]
+
     # Node (Memory) Management
     @abstractmethod
     def add_node(self, id: str, memory: str, metadata: dict[str, Any]) -> None:
@@ -144,16 +167,23 @@ def get_context_chain(self, id: str, type: str = "FOLLOWS") -> list[str]:
 
     # Search / recall operations
     @abstractmethod
-    def search_by_embedding(self, vector: list[float], top_k: int = 5, **kwargs) -> list[dict]:
+    def search_by_embedding(
+        self, vector: list[float], top_k: int = 5, return_fields: list[str] | None = None, **kwargs
+    ) -> list[dict]:
         """
         Retrieve node IDs based on vector similarity.
 
         Args:
             vector (list[float]): The embedding vector representing query semantics.
             top_k (int): Number of top similar nodes to retrieve.
+            return_fields (list[str], optional): Additional node fields to include in results
+                (e.g., ["memory", "status", "tags"]). When provided, each result dict will
+                contain these fields in addition to 'id' and 'score'.
+                Defaults to None (only 'id' and 'score' are returned).
 
         Returns:
             list[dict]: A list of dicts with 'id' and 'score', ordered by similarity.
+                If return_fields is specified, each dict also includes the requested fields.
 
         Notes:
             - This method may internally call a VecDB (e.g., Qdrant) or store embeddings in the graph DB itself.

src/memos/graph_dbs/neo4j.py

@@ -818,6 +818,7 @@ def search_by_embedding(
         user_name: str | None = None,
         filter: dict | None = None,
         knowledgebase_ids: list[str] | None = None,
+        return_fields: list[str] | None = None,
         **kwargs,
     ) -> list[dict]:
         """
@@ -832,9 +833,14 @@ def search_by_embedding(
             threshold (float, optional): Minimum similarity score threshold (0 ~ 1).
             search_filter (dict, optional): Additional metadata filters for search results.
                             Keys should match node properties, values are the expected values.
+            return_fields (list[str], optional): Additional node fields to include in results
+                            (e.g., ["memory", "status", "tags"]). When provided, each result
+                            dict will contain these fields in addition to 'id' and 'score'.
+                            Defaults to None (only 'id' and 'score' are returned).
 
         Returns:
             list[dict]: A list of dicts with 'id' and 'score', ordered by similarity.
+                If return_fields is specified, each dict also includes the requested fields.
 
         Notes:
             - This method uses Neo4j native vector indexing to search for similar nodes.
@@ -886,11 +892,20 @@ def search_by_embedding(
         if where_clauses:
             where_clause = "WHERE " + " AND ".join(where_clauses)
 
+        return_clause = "RETURN node.id AS id, score"
+        if return_fields:
+            validated_fields = self._validate_return_fields(return_fields)
+            extra_fields = ", ".join(
+                f"node.{field} AS {field}" for field in validated_fields if field != "id"
+            )
+            if extra_fields:
+                return_clause = f"RETURN node.id AS id, score, {extra_fields}"
+
         query = f"""
             CALL db.index.vector.queryNodes('memory_vector_index', $k, $embedding)
             YIELD node, score
             {where_clause}
-            RETURN node.id AS id, score
+            {return_clause}
         """
 
         parameters = {"embedding": vector, "k": top_k}
@@ -920,7 +935,15 @@ def search_by_embedding(
         print(f"[search_by_embedding] query: {query},parameters: {parameters}")
         with self.driver.session(database=self.db_name) as session:
             result = session.run(query, parameters)
-            records = [{"id": record["id"], "score": record["score"]} for record in result]
+            records = []
+            for record in result:
+                item = {"id": record["id"], "score": record["score"]}
+                if return_fields:
+                    record_keys = record.keys()
+                    for field in return_fields:
+                        if field != "id" and field in record_keys:
+                            item[field] = record[field]
+                records.append(item)
 
         # Threshold filtering after retrieval
         if threshold is not None:
@@ -943,8 +966,8 @@ def search_by_fulltext(
         **kwargs,
     ) -> list[dict]:
         """
-        TODO: 实现 Neo4j 的关键词检索, 以兼容 TreeTextMemory 的 keyword/fulltext 召回路径.
-        目前先返回空列表, 避免切换到 Neo4j 后因缺失方法导致运行时报错.
+        TODO: Implement fulltext search for Neo4j to be compatible with TreeTextMemory's keyword/fulltext recall path.
+        Currently, return an empty list to avoid runtime errors due to missing methods when switching to Neo4j.
         """
         return []