Merge pull request #112 from dot-agent/fixing-memory

Fixing memory

nextpy/ai/engine/_program.py

@@ -154,6 +154,7 @@ def __init__(
         log=None,
         memory=None,
         memory_threshold=1,
+        memory_llm=None,
         **kwargs,
     ):
         """Create a new Program object from a program string.
@@ -222,19 +223,19 @@ def __init__(
         self.await_missing = await_missing
         self.log = log
         self.memory = memory
-        self.memory_threshold = memory_threshold
 
         if self.memory is not None:
             if not isinstance(self.memory, BaseMemory):
                 raise TypeError("Memory type is not compatible")
 
+            # If user has not passed separate llm for memory, use engine's
+            self.memory.llm = memory_llm if memory_llm is not None else llm
+            self.memory.memory_threshold = memory_threshold
+
             if self._text.find("ConversationHistory") == -1:
                 self._text = add_variable(self._text)
 
-        if self.memory is not None:
-            ConversationHistory = self.memory.get_memory(
-                memory_threshold=self.memory_threshold
-            )
+            ConversationHistory = self.memory.get_memory()
             kwargs["ConversationHistory"] = ConversationHistory
             self.ConversationHistory = ConversationHistory
 
@@ -382,9 +383,7 @@ def __call__(self, from_agent=False, **kwargs):
 
         if self.memory is not None:
             if not from_agent:
-                self.ConversationHistory = self.memory.get_memory(
-                    memory_threshold=self.memory_threshold
-                )
+                self.ConversationHistory = self.memory.get_memory()
             kwargs["ConversationHistory"] = self.ConversationHistory
 
         log.debug(f"in __call__ with kwargs: {kwargs}")

nextpy/ai/memory/base.py

@@ -14,6 +14,8 @@ class BaseMemory(ABC, BaseModel):
     This class defines the interface for setting, getting, and checking existence of data in memory.
     """
 
+    llm: Any = None
+    memory_threshold: int = 1
     # All memories are stored in this list
     messages: List[Dict[BaseMessage, Any]] = Field(default=list())
 

nextpy/ai/memory/read_only.py

@@ -11,9 +11,6 @@ class ReadOnlyMemory(BaseMemory):
 
     memory: BaseMemory
 
-    def __init__(self, memory: BaseMemory):
-        self.memory = memory
-
     def add_memory(self, prompt: str, llm_response: Any) -> None:
         """cannot edit a read only memory."""
         pass

nextpy/ai/memory/summary.py

@@ -40,7 +40,7 @@ def add_memory(self, prompt: str, llm_response: Any) -> None:
 
     def get_memory(self, **kwargs) -> str:
         """Retrieve entire memory from the store."""
-        # Create llm instance
+
         llm = engine.llms.OpenAI(model="gpt-3.5-turbo")
 
         new_messages = [
@@ -59,10 +59,9 @@ def get_memory(self, **kwargs) -> str:
                     + "\n"
                 )
                 self.messages_in_summary.append(conversation)
-
-            summarizer = engine(template=SUMMARIZER_TEMPLATE, llm=llm, stream=False)
+            summarizer = engine(template=SUMMARIZER_TEMPLATE, stream=False, llm=self.llm if self.llm is not None else llm)
             summarized_memory = summarizer(
-                summary=self.current_summary, new_lines=messages_to_text
+                summary=self.current_summary, new_lines=messages_to_text,
             )
             self.current_summary = extract_text(summarized_memory.text)
             summarized_memory = "Current conversation:\n" + self.current_summary
@@ -96,7 +95,7 @@ def remove_memory(self, prompt: str, llm=Any) -> None:
                         self.messages_in_summary.append(conversation)
 
                     summarizer = engine(
-                        template=SUMMARIZER_TEMPLATE, llm=llm, stream=False
+                        template=SUMMARIZER_TEMPLATE, llm=self.llm, stream=False
                     )
                     summarized_memory = summarizer(
                         summary="", new_lines=messages_to_text