Merge pull request #2 from memfuse/dev

Dev

Author: M1n9X

examples/04_gradio_chatbot.py

@@ -8,6 +8,13 @@
 
 # Global username variable
 USERNAME = "Jane Doe"
+SYSTEM_MESSAGE = (
+    "You are a helpful AI assistant with access to a persistent long-term memory. "
+    "You can recall, reference, and use information from previous conversations with the user. "
+    "Leverage this memory to provide more relevant, helpful, and context-aware answers. "
+    "If you remember something from earlier, feel free to mention it. "
+    "If the user mentions something from a previous interaction that you don't remember, please apologize and say you must have forgotten."
+)
 
 def main():
     # Make MemFuse base URL configurable via environment variable
@@ -49,8 +56,12 @@ def memfuse_chatbot(message, history):
                 for i, item in enumerate(history):
                     print(f"DEBUG: History item {i}: type={type(item)}, content={item}")
                     if isinstance(item, dict):
-                        # History is already in message format
-                        messages_history.append(item)
+                        # History is already in message format - extract only role and content
+                        if 'role' in item and 'content' in item:
+                            messages_history.append({
+                                "role": item["role"],
+                                "content": item["content"]
+                            })
                     elif isinstance(item, (list, tuple)):
                         # Handle tuple/list format - could be (user_msg, assistant_msg) or more items
                         if len(item) >= 2:
@@ -66,7 +77,7 @@ def memfuse_chatbot(message, history):
                         print(f"Unknown history item format: {type(item)}: {item}")
 
             print(f"DEBUG: Final messages_history: {messages_history}")
-            current_messages_for_api = messages_history + [{"role": "user", "content": message}]
+            current_messages_for_api = [{"role": "system", "content": SYSTEM_MESSAGE}] + messages_history + [{"role": "user", "content": message}]
             print(f"DEBUG: Sending to API: {current_messages_for_api}")
 
             try:

poetry.lock

@@ -44,10 +44,11 @@ ipykernel = "^6.29.5"
 datasets = "^3.6.0"
 colorlog = "^6.9.0"
 twine = "^6.1.0"
+poetry-dynamic-versioning = "^1.8.2"
 
 [build-system]
-requires = ["poetry-core"]
-build-backend = "poetry.core.masonry.api"
+requires = ["poetry-core", "poetry-dynamic-versioning"]
+build-backend = "poetry_dynamic_versioning.backend"
 
 [tool.poetry-dynamic-versioning]
 enable = true
@@ -56,6 +57,7 @@ style = "pep440"
 
 [tool.poetry-dynamic-versioning.substitution]
 files = ["src/memfuse/__init__.py"]
+patterns = ["(^__version__\\s*(?::.*?)?\\s*=\\s*['\"])[^'\"]*(['\"])"]
 
 [tool.pytest.ini_options]
 markers = [

pyproject.toml

@@ -1,6 +1,6 @@
 """MemFuse Python Client Library"""
 
-__version__ = "{{version}}"  # Or your actual version
+__version__ = "{{version}}" 
 
 from .client import AsyncMemFuse, MemFuse
 from .memory import AsyncMemory, Memory

src/memfuse/__init__.py

@@ -33,11 +33,12 @@ def _get_method(self, endpoint_key: str) -> str:
         return self.ENDPOINTS[endpoint_key]['method']
 
     def _build_list_url(
-        self, 
-        session_id: str, 
+        self,
+        session_id: str,
         limit: Optional[int] = None,
         sort_by: Optional[str] = None,
-        order: Optional[str] = None
+        order: Optional[str] = None,
+        buffer_only: Optional[bool] = None
     ) -> str:
         """Build URL for list endpoint with optional query parameters."""
         query_params = []
@@ -47,7 +48,9 @@ def _build_list_url(
             query_params.append(f"sort_by={sort_by}")
         if order is not None:
             query_params.append(f"order={order}")
-        
+        if buffer_only is not None:
+            query_params.append(f"buffer_only={str(buffer_only).lower()}")
+
         query_string = "&".join(query_params)
         endpoint = self._build_url('list', session_id=session_id)
         if query_string:
@@ -61,6 +64,7 @@ async def list(
         limit: Optional[int] = 20,
         sort_by: Optional[str] = "timestamp",
         order: Optional[str] = "desc",
+        buffer_only: Optional[bool] = None,
     ) -> Dict[str, Any]:
         """List all messages in a session.
 
@@ -69,11 +73,12 @@ async def list(
             limit: Maximum number of messages to return. Defaults to 20.
             sort_by: Field to sort messages by (e.g., "timestamp", "id"). Defaults to "timestamp".
             order: Sort order ("asc" or "desc"). Defaults to "desc".
+            buffer_only: If True, only return RoundBuffer data; if False, return HybridBuffer + SQLite data excluding RoundBuffer
 
         Returns:
             Response data
         """
-        url = self._build_list_url(session_id, limit, sort_by, order)
+        url = self._build_list_url(session_id, limit, sort_by, order, buffer_only)
         return await self.client._request(self._get_method('list'), url)
 
     async def add(self, session_id: str, messages: List[Dict[str, str]]) -> Dict[str, Any]:
@@ -159,6 +164,7 @@ def list_sync(
         limit: Optional[int] = 20,
         sort_by: Optional[str] = "timestamp",
         order: Optional[str] = "desc",
+        buffer_only: Optional[bool] = None,
     ) -> Dict[str, Any]:
         """List all messages in a session (sync version).
 
@@ -167,11 +173,12 @@ def list_sync(
             limit: Maximum number of messages to return. Defaults to 20.
             sort_by: Field to sort messages by (e.g., "timestamp", "id"). Defaults to "timestamp".
             order: Sort order ("asc" or "desc"). Defaults to "desc".
+            buffer_only: If True, only return RoundBuffer data; if False, return HybridBuffer + SQLite data excluding RoundBuffer
 
         Returns:
             Response data
         """
-        url = self._build_list_url(session_id, limit, sort_by, order)
+        url = self._build_list_url(session_id, limit, sort_by, order, buffer_only)
         return self.client._request_sync(self._get_method('list'), url)
 
     def add_sync(self, session_id: str, messages: List[Dict[str, str]]) -> Dict[str, Any]:

src/memfuse/api/messages.py

@@ -102,7 +102,7 @@ async def _request(
                     f"Cannot connect to MemFuse server at {self.base_url}. "
                     "Please make sure the server is running.\n\n"
                     "You can start the server with:\n"
-                    "  python -m memfuse.server --host localhost --port 8000"
+                    "  poetry run memfuse-core"
                 )
 
             url = f"{self.base_url}{endpoint}"
@@ -121,7 +121,7 @@ async def _request(
                 f"Cannot connect to MemFuse server at {self.base_url}. "
                 "Please make sure the server is running.\n\n"
                 "You can start the server with:\n"
-                "  python -m memfuse.server --host localhost --port 8000"
+                "  poetry run memfuse-core"
             ) from e
 
     async def init(
@@ -320,7 +320,7 @@ def _request_sync(
                     f"Cannot connect to MemFuse server at {self.base_url}. "
                     "Please make sure the server is running.\n\n"
                     "You can start the server with:\n"
-                    "  python -m memfuse.server --host localhost --port 8000"
+                    "  poetry run memfuse-core"
                 )
 
             url = f"{self.base_url}{endpoint}"
@@ -337,7 +337,7 @@ def _request_sync(
                 f"Cannot connect to MemFuse server at {self.base_url}. "
                 "Please make sure the server is running.\n\n"
                 "You can start the server with:\n"
-                "  python -m memfuse.server --host localhost --port 8000"
+                "  poetry run memfuse-core"
             ) from e
 
     def init(

src/memfuse/client.py

@@ -48,11 +48,22 @@ def wrapper(*args: Any, **kwargs: Any) -> Any:  # signature replaced below
         # ------- 2. Get the last n messages ----------------------------------
         max_chat_history = memory.max_chat_history
 
-        retrieved_chat_history = memory.list_messages(
+        in_buffer_chat_history = memory.list_messages(
             limit=max_chat_history,
+            buffer_only=True,
         )
 
-        chat_history = [{"role": message["role"], "content": message["content"]} for message in retrieved_chat_history["data"]["messages"][::-1]]
+        in_buffer_messages_length = len(in_buffer_chat_history["data"]["messages"])
+
+        if in_buffer_messages_length < max_chat_history:
+            in_db_chat_history = memory.list_messages(
+                limit=max_chat_history - in_buffer_messages_length,
+                buffer_only=False,
+            )
+        else:
+            in_db_chat_history = []
+
+        chat_history = [{"role": message["role"], "content": message["content"]} for message in in_db_chat_history["data"]["messages"][::-1]] + [{"role": message["role"], "content": message["content"]} for message in in_buffer_chat_history["data"]["messages"][::-1]]
 
         # ------- 3. Retrieve memories ---------------------------------------
         # Convert Anthropic formatted messages to a string for querying
@@ -146,11 +157,22 @@ async def wrapper(*args: Any, **kwargs: Any) -> Any:  # signature replaced below
         # ------- 2. Get the last n messages ----------------------------------
         max_chat_history = memory.max_chat_history
 
-        retrieved_chat_history = await memory.list_messages(
+        in_buffer_chat_history = await memory.list_messages(
             limit=max_chat_history,
+            buffer_only=True,
         )
 
-        chat_history = [{"role": message["role"], "content": message["content"]} for message in retrieved_chat_history["data"]["messages"][::-1]]
+        in_buffer_messages_length = len(in_buffer_chat_history["data"]["messages"])
+
+        if in_buffer_messages_length < max_chat_history:
+            in_db_chat_history = await memory.list_messages(
+                limit=max_chat_history - in_buffer_messages_length,
+                buffer_only=False,
+            )
+        else:
+            in_db_chat_history = []
+
+        chat_history = [{"role": message["role"], "content": message["content"]} for message in in_db_chat_history["data"]["messages"][::-1]] + [{"role": message["role"], "content": message["content"]} for message in in_buffer_chat_history["data"]["messages"][::-1]]
 
         # ------- 3. Retrieve memories ---------------------------------------
         query_string = PromptFormatter.messages_to_query(chat_history + query_messages)

src/memfuse/llm/anthropic_adapter.py

@@ -10,7 +10,7 @@
 # from google.genai import AsyncClient as AsyncGeminiClient # google.genai.Client can be used with an async transport
 
 from memfuse import Memory
-from memfuse.prompts import PromptContext
+from memfuse.prompts import PromptContext, PromptFormatter
 
 # Set up logger for this module
 logger = logging.getLogger(__name__)
@@ -133,26 +133,41 @@ def _instrument_generate_content_sync(
         latest_user_query_message = gemini_query_messages[-1]
 
     retrieved_memories = None
-    retrieved_chat_history = None
+    chat_history = None
 
-    if latest_user_query_message:
-        query_response = memory.query_session(latest_user_query_message["content"])
-        retrieved_memories = query_response["data"]["results"] if query_response else None
-        
+    if latest_user_query_message:    
         # Get chat history
         max_chat_history = memory.max_chat_history
-        chat_history_response = memory.list_messages(limit=max_chat_history)
-        if chat_history_response and chat_history_response.get("data", {}).get("messages"):
-            retrieved_chat_history = [
-                {"role": msg["role"], "content": msg["content"]} 
-                for msg in chat_history_response["data"]["messages"][::-1]
-            ]
+
+        in_buffer_chat_history = memory.list_messages(
+            limit=max_chat_history,
+            buffer_only=True,
+        )
+
+        in_buffer_messages_length = len(in_buffer_chat_history["data"]["messages"])
+
+        if in_buffer_messages_length < max_chat_history:
+            in_db_chat_history = memory.list_messages(
+                limit=max_chat_history - in_buffer_messages_length,
+                buffer_only=False,
+            )
+        else:
+            in_db_chat_history = []
+
+        chat_history = [{"role": message["role"], "content": message["content"]} for message in in_db_chat_history["data"]["messages"][::-1]] + [{"role": message["role"], "content": message["content"]} for message in in_buffer_chat_history["data"]["messages"][::-1]]
+
+        # Retrieve memories
+        query_string = PromptFormatter.messages_to_query(chat_history + gemini_query_messages)
+        query_response = memory.query_session(query_string)
+        retrieved_memories = query_response["data"]["results"] if query_response else None
+
+        logger.info(f"Retrieved memories: {retrieved_memories}")
 
     # 3. Compose the prompt context for PromptFormatter
     prompt_context = PromptContext(
         query_messages=gemini_query_messages,
         retrieved_memories=retrieved_memories,
-        retrieved_chat_history=retrieved_chat_history,
+        retrieved_chat_history=chat_history,
         max_chat_history=memory.max_chat_history,
     )
 
@@ -191,19 +206,33 @@ async def _instrument_generate_content_async(
     retrieved_memories = None
     retrieved_chat_history = None
 
-    if latest_user_query_message:
-        # Properly await async memory operations
-        query_response = await memory.query_session(latest_user_query_message["content"])
-        retrieved_memories = query_response["data"]["results"] if query_response else None
-        
+    if latest_user_query_message:    
         # Get chat history
         max_chat_history = memory.max_chat_history
-        chat_history_response = await memory.list_messages(limit=max_chat_history)
-        if chat_history_response and chat_history_response.get("data", {}).get("messages"):
-            retrieved_chat_history = [
-                {"role": msg["role"], "content": msg["content"]} 
-                for msg in chat_history_response["data"]["messages"][::-1]
-            ]
+
+        in_buffer_chat_history = await memory.list_messages(
+            limit=max_chat_history,
+            buffer_only=True,
+        )
+
+        in_buffer_messages_length = len(in_buffer_chat_history["data"]["messages"])
+
+        if in_buffer_messages_length < max_chat_history:
+            in_db_chat_history = await memory.list_messages(
+                limit=max_chat_history - in_buffer_messages_length,
+                buffer_only=False,
+            )
+        else:
+            in_db_chat_history = []
+
+        retrieved_chat_history = [{"role": message["role"], "content": message["content"]} for message in in_db_chat_history["data"]["messages"][::-1]] + [{"role": message["role"], "content": message["content"]} for message in in_buffer_chat_history["data"]["messages"][::-1]]
+
+        # Retrieve memories
+        query_string = PromptFormatter.messages_to_query(retrieved_chat_history + gemini_query_messages)
+        query_response = await memory.query_session(query_string)
+        retrieved_memories = query_response["data"]["results"] if query_response else None
+
+        logger.info(f"Retrieved memories: {retrieved_memories}")
 
     # 3. Compose the prompt context for PromptFormatter
     prompt_context = PromptContext(