Merge branch 'dev_new' into feat/deep-search

Author: fridayL

docker/requirements.txt

@@ -160,3 +160,4 @@ xlrd==2.0.2
 xlsxwriter==3.2.5
 prometheus-client==0.23.1
 pymilvus==2.5.12
+langchain-text-splitters==1.0.0

src/memos/api/handlers/add_handler.py

@@ -5,6 +5,8 @@
 using dependency injection for better modularity and testability.
 """
 
+from pydantic import validate_call
+
 from memos.api.handlers.base_handler import BaseHandler, HandlerDependencies
 from memos.api.product_models import APIADDRequest, APIFeedbackRequest, MemoryResponse
 from memos.memories.textual.item import (
@@ -13,6 +15,7 @@
 from memos.multi_mem_cube.composite_cube import CompositeCubeView
 from memos.multi_mem_cube.single_cube import SingleCubeView
 from memos.multi_mem_cube.views import MemCubeView
+from memos.types import MessageList
 
 
 class AddHandler(BaseHandler):
@@ -60,38 +63,45 @@ def handle_add_memories(self, add_req: APIADDRequest) -> MemoryResponse:
 
         cube_view = self._build_cube_view(add_req)
 
+        @validate_call
+        def _check_messages(messages: MessageList) -> None:
+            pass
+
         if add_req.is_feedback:
-            chat_history = add_req.chat_history
-            messages = add_req.messages
-            if chat_history is None:
-                chat_history = []
-            if messages is None:
-                messages = []
-            concatenate_chat = chat_history + messages
-
-            last_user_index = max(i for i, d in enumerate(concatenate_chat) if d["role"] == "user")
-            feedback_content = concatenate_chat[last_user_index]["content"]
-            feedback_history = concatenate_chat[:last_user_index]
-
-            feedback_req = APIFeedbackRequest(
-                user_id=add_req.user_id,
-                session_id=add_req.session_id,
-                task_id=add_req.task_id,
-                history=feedback_history,
-                feedback_content=feedback_content,
-                writable_cube_ids=add_req.writable_cube_ids,
-                async_mode=add_req.async_mode,
-            )
-            process_record = cube_view.feedback_memories(feedback_req)
+            try:
+                messages = add_req.messages
+                _check_messages(messages)
 
-            self.logger.info(
-                f"[FeedbackHandler] Final feedback results count={len(process_record)}"
-            )
+                chat_history = add_req.chat_history if add_req.chat_history else []
+                concatenate_chat = chat_history + messages
 
-            return MemoryResponse(
-                message="Memory feedback successfully",
-                data=[process_record],
-            )
+                last_user_index = max(
+                    i for i, d in enumerate(concatenate_chat) if d["role"] == "user"
+                )
+                feedback_content = concatenate_chat[last_user_index]["content"]
+                feedback_history = concatenate_chat[:last_user_index]
+
+                feedback_req = APIFeedbackRequest(
+                    user_id=add_req.user_id,
+                    session_id=add_req.session_id,
+                    task_id=add_req.task_id,
+                    history=feedback_history,
+                    feedback_content=feedback_content,
+                    writable_cube_ids=add_req.writable_cube_ids,
+                    async_mode=add_req.async_mode,
+                )
+                process_record = cube_view.feedback_memories(feedback_req)
+
+                self.logger.info(
+                    f"[ADDFeedbackHandler] Final feedback results count={len(process_record)}"
+                )
+
+                return MemoryResponse(
+                    message="Memory feedback successfully",
+                    data=[process_record],
+                )
+            except Exception as e:
+                self.logger.warning(f"[ADDFeedbackHandler] Running error: {e}")
 
         results = cube_view.add_memories(add_req)
 

src/memos/api/handlers/chat_handler.py

@@ -142,7 +142,9 @@ def handle_chat_complete(self, chat_req: APIChatCompleteRequest) -> dict[str, An
 
             # Step 2: Build system prompt
             system_prompt = self._build_system_prompt(
-                filtered_memories, search_response.data["pref_string"], chat_req.system_prompt
+                filtered_memories,
+                search_response.data.get("pref_string", ""),
+                chat_req.system_prompt,
             )
 
             # Prepare message history
@@ -257,7 +259,7 @@ def generate_chat_response() -> Generator[str, None, None]:
                     # Step 2: Build system prompt with memories
                     system_prompt = self._build_system_prompt(
                         filtered_memories,
-                        search_response.data["pref_string"],
+                        search_response.data.get("pref_string", ""),
                         chat_req.system_prompt,
                     )
 
@@ -449,7 +451,7 @@ def generate_chat_response() -> Generator[str, None, None]:
 
                     # Step 2: Build system prompt with memories
                     system_prompt = self._build_enhance_system_prompt(
-                        filtered_memories, search_response.data["pref_string"]
+                        filtered_memories, search_response.data.get("pref_string", "")
                     )
 
                     # Prepare messages

src/memos/api/handlers/formatters_handler.py

@@ -90,3 +90,37 @@ def post_process_pref_mem(
         memories_result["pref_note"] = pref_note
 
     return memories_result
+
+
+def post_process_textual_mem(
+    memories_result: dict[str, Any],
+    text_formatted_mem: list[dict[str, Any]],
+    mem_cube_id: str,
+) -> dict[str, Any]:
+    """
+    Post-process text and tool memory results.
+    """
+    fact_mem = [
+        mem
+        for mem in text_formatted_mem
+        if mem["metadata"]["memory_type"] not in ["ToolSchemaMemory", "ToolTrajectoryMemory"]
+    ]
+    tool_mem = [
+        mem
+        for mem in text_formatted_mem
+        if mem["metadata"]["memory_type"] in ["ToolSchemaMemory", "ToolTrajectoryMemory"]
+    ]
+
+    memories_result["text_mem"].append(
+        {
+            "cube_id": mem_cube_id,
+            "memories": fact_mem,
+        }
+    )
+    memories_result["tool_mem"].append(
+        {
+            "cube_id": mem_cube_id,
+            "memories": tool_mem,
+        }
+    )
+    return memories_result

src/memos/api/product_models.py

@@ -6,7 +6,7 @@
 
 # Import message types from core types module
 from memos.log import get_logger
-from memos.types import MessageDict, PermissionDict, SearchMode
+from memos.types import MessageList, MessagesType, PermissionDict, SearchMode
 
 
 logger = get_logger(__name__)
@@ -56,7 +56,7 @@ class Message(BaseModel):
 
 class MemoryCreate(BaseRequest):
     user_id: str = Field(..., description="User ID")
-    messages: list | None = Field(None, description="List of messages to store.")
+    messages: MessageList | None = Field(None, description="List of messages to store.")
     memory_content: str | None = Field(None, description="Content to store as memory")
     doc_path: str | None = Field(None, description="Path to document to store")
     mem_cube_id: str | None = Field(None, description="ID of the memory cube")
@@ -83,7 +83,7 @@ class ChatRequest(BaseRequest):
     writable_cube_ids: list[str] | None = Field(
         None, description="List of cube IDs user can write for multi-cube chat"
     )
-    history: list | None = Field(None, description="Chat history")
+    history: MessageList | None = Field(None, description="Chat history")
     mode: SearchMode = Field(SearchMode.FAST, description="search mode: fast, fine, or mixture")
     system_prompt: str | None = Field(None, description="Base system prompt to use for chat")
     top_k: int = Field(10, description="Number of results to return")
@@ -165,7 +165,7 @@ class ChatCompleteRequest(BaseRequest):
     user_id: str = Field(..., description="User ID")
     query: str = Field(..., description="Chat query message")
     mem_cube_id: str | None = Field(None, description="Cube ID to use for chat")
-    history: list | None = Field(None, description="Chat history")
+    history: MessageList | None = Field(None, description="Chat history")
     internet_search: bool = Field(False, description="Whether to use internet search")
     system_prompt: str | None = Field(None, description="Base prompt to use for chat")
     top_k: int = Field(10, description="Number of results to return")
@@ -251,7 +251,7 @@ class MemoryCreateRequest(BaseRequest):
     """Request model for creating memories."""
 
     user_id: str = Field(..., description="User ID")
-    messages: str | list | None = Field(None, description="List of messages to store.")
+    messages: str | MessagesType | None = Field(None, description="List of messages to store.")
     memory_content: str | None = Field(None, description="Memory content to store")
     doc_path: str | None = Field(None, description="Path to document to store")
     mem_cube_id: str | None = Field(None, description="Cube ID")
@@ -326,6 +326,21 @@ class APISearchRequest(BaseRequest):
         ),
     )
 
+    search_tool_memory: bool = Field(
+        True,
+        description=(
+            "Whether to retrieve tool memories along with general memories. "
+            "If enabled, the system will automatically recall tool memories "
+            "relevant to the query. Default: True."
+        ),
+    )
+
+    tool_mem_top_k: int = Field(
+        6,
+        ge=0,
+        description="Number of tool memories to retrieve (top-K). Default: 6.",
+    )
+
     # ==== Filter conditions ====
     # TODO: maybe add detailed description later
     filter: dict[str, Any] | None = Field(
@@ -360,7 +375,7 @@ class APISearchRequest(BaseRequest):
     )
 
     # ==== Context ====
-    chat_history: list | None = Field(
+    chat_history: MessageList | None = Field(
         None,
         description=(
             "Historical chat messages used internally by algorithms. "
@@ -490,7 +505,7 @@ class APIADDRequest(BaseRequest):
     )
 
     # ==== Input content ====
-    messages: str | list | None = Field(
+    messages: MessagesType | None = Field(
         None,
         description=(
             "List of messages to store. Supports: "
@@ -506,7 +521,7 @@ class APIADDRequest(BaseRequest):
     )
 
     # ==== Chat history ====
-    chat_history: list | None = Field(
+    chat_history: MessageList | None = Field(
         None,
         description=(
             "Historical chat messages used internally by algorithms. "
@@ -636,21 +651,20 @@ class APIFeedbackRequest(BaseRequest):
         "default_session", description="Session ID for soft-filtering memories"
     )
     task_id: str | None = Field(None, description="Task ID for monitering async tasks")
-    history: list[MessageDict] | None = Field(..., description="Chat history")
+    history: MessageList | None = Field(..., description="Chat history")
     retrieved_memory_ids: list[str] | None = Field(
         None, description="Retrieved memory ids at last turn"
     )
     feedback_content: str | None = Field(..., description="Feedback content to process")
     feedback_time: str | None = Field(None, description="Feedback time")
-    # ==== Multi-cube writing ====
     writable_cube_ids: list[str] | None = Field(
         None, description="List of cube IDs user can write for multi-cube add"
     )
     async_mode: Literal["sync", "async"] = Field(
         "async", description="feedback mode: sync or async"
     )
     corrected_answer: bool = Field(False, description="Whether need return corrected answer")
-    # ==== Backward compatibility ====
+    # ==== mem_cube_id is NOT enabled====
     mem_cube_id: str | None = Field(
         None,
         description=(
@@ -671,7 +685,7 @@ class APIChatCompleteRequest(BaseRequest):
     writable_cube_ids: list[str] | None = Field(
         None, description="List of cube IDs user can write for multi-cube chat"
     )
-    history: list | None = Field(None, description="Chat history")
+    history: MessageList | None = Field(None, description="Chat history")
     mode: SearchMode = Field(SearchMode.FAST, description="search mode: fast, fine, or mixture")
     system_prompt: str | None = Field(None, description="Base system prompt to use for chat")
     top_k: int = Field(10, description="Number of results to return")
@@ -740,7 +754,7 @@ class SuggestionRequest(BaseRequest):
     user_id: str = Field(..., description="User ID")
     mem_cube_id: str = Field(..., description="Cube ID")
     language: Literal["zh", "en"] = Field("zh", description="Language for suggestions")
-    message: list | None = Field(None, description="List of messages to store.")
+    message: MessagesType | None = Field(None, description="List of messages to store.")
 
 
 # ─── MemOS Client Response Models ──────────────────────────────────────────────

src/memos/configs/embedder.py

@@ -12,6 +12,10 @@ class BaseEmbedderConfig(BaseConfig):
     embedding_dims: int | None = Field(
         default=None, description="Number of dimensions for the embedding"
     )
+    max_tokens: int | None = Field(
+        default=8192,
+        description="Maximum number of tokens per text. Texts exceeding this limit will be automatically truncated. Set to None to disable truncation.",
+    )
     headers_extra: dict[str, Any] | None = Field(
         default=None,
         description="Extra headers for the embedding model, only for universal_api backend",

src/memos/embedders/ark.py

@@ -49,6 +49,9 @@ def embed(self, texts: list[str]) -> list[list[float]]:
             MultimodalEmbeddingContentPartTextParam,
         )
 
+        # Truncate texts if max_tokens is configured
+        texts = self._truncate_texts(texts)
+
         if self.config.multi_modal:
             texts_input = [
                 MultimodalEmbeddingContentPartTextParam(text=text, type="text") for text in texts