delayed memory recall

README.md

@@ -176,6 +176,7 @@ docker run -p 50001:80 agent0ai/agent-zero
 - New notifications system
 - New local terminal interface for stability
 - Rate limiter integration to models
+- Delayed memory recall
 - Smarter autoscrolling in UI
 - Action buttons in messages
 - Multiple API keys support

prompts/memory.memories_query.sys.md

@@ -7,11 +7,18 @@
 - The response format is a plain text string containing the query
 - No other text, no formatting
 
+# No query
+- If the conversation is not relevant for memory search, return a single dash (-)
+
 # Rules
 - Only focus on facts and events, ignore common conversation patterns, greeting etc.
 - Ignore AI thoughts and behavior
 - Focus on USER MESSAGE if provided, use HISTORY for context
 
+# Ignored:
+For the following topics, no query is needed and return a single dash (-):
+- Greeting
+
 # Example
 ```json
 USER: "Write a song about my dog"

prompts/memory.memories_sum.sys.md

@@ -36,10 +36,18 @@
 > AsyncRaceError in primary_modules.py was fixed by adding a thread lock on line 123 (important event with details for context)
 > Local SQL database was created, server is running on port 3306 (important event with details for context)
 
-# Wrong examples with (explanation of error), never output memories like these 
+# WRONG examples with (explanation of error), never output memories like these 
 > Dog Information (no useful facts)
 > User greeted with 'hi' (just conversation, not useful in the future )
 > Respond with a warm greeting and invite further conversation (do not memorize AI's instructions or thoughts)
 > User's name (details missing, not useful)
 > Today is Monday (just date, no value in this information)
-> Market inquiry (just a topic without detail)
+> Market inquiry (just a topic without detail)
+> RAM Status (just a topic without detail)
+> The user requested current RAM and CPU status. (No exact facts to memorize)
+
+
+# Further WRONG examples
+- Hello
+- The user requested current RAM and CPU status.
+- 

prompts/memory.recall_delay_msg.md

@@ -0,0 +1 @@
+Info: auto memory recall set to delayed mode. auto memories will be available after next message. if manual memory check is required use memory tools.

python/extensions/message_loop_prompts_after/_50_recall_memories.py

@@ -3,9 +3,11 @@ from python.helpers.extension import Extension
 from python.helpers.memory import Memory
 from agent import LoopData
 from python.tools.memory_load import DEFAULT_THRESHOLD as DEFAULT_MEMORY_THRESHOLD
-from python.helpers import dirty_json, errors, settings
+from python.helpers import dirty_json, errors, settings, log 
+
 
 DATA_NAME_TASK = "_recall_memories_task"
+DATA_NAME_ITER = "_recall_memories_iter"
 
 
 class RecallMemories(Extension):
@@ -22,18 +24,30 @@ class RecallMemories(Extension):
 
         set = settings.get_settings()
 
-        # every 3 iterations (or the first one) recall memories
+        # turned off in settings?
+        if not set["memory_recall_enabled"]:
+            return
+
+        # every X iterations (or the first one) recall memories
         if loop_data.iteration % set["memory_recall_interval"] == 0:
+
+            # show util message right away
+            log_item = self.agent.context.log.log(
+                type="util",
+                heading="Searching memories...",
+            )
+
             task = asyncio.create_task(
-                self.search_memories(loop_data=loop_data, **kwargs)
+                self.search_memories(loop_data=loop_data, log_item=log_item, **kwargs)
             )
         else:
             task = None
 
         # set to agent to be able to wait for it
         self.agent.set_data(DATA_NAME_TASK, task)
+        self.agent.set_data(DATA_NAME_ITER, loop_data.iteration)
 
-    async def search_memories(self, loop_data: LoopData, **kwargs):
+    async def search_memories(self, log_item: log.LogItem, loop_data: LoopData, **kwargs):
 
         # cleanup
         extras = loop_data.extras_persistent
@@ -46,16 +60,6 @@ class RecallMemories(Extension):
         set = settings.get_settings()
         # try:
 
-        # if recall is disabled, return
-        if not set["memory_recall_enabled"]:
-            return
-
-        # show full util message
-        log_item = self.agent.context.log.log(
-            type="util",
-            heading="Searching memories...",
-        )
-
         # get system message and chat history for util llm
         system = self.agent.read_prompt("memory.memories_query.sys.md")
 
@@ -100,6 +104,13 @@ class RecallMemories(Extension):
         else:
             query = user_instruction + "\n\n" + history
 
+        # if there is no query (or just dash by the LLM), do not continue
+        if not query or len(query) <= 3:
+            log_item.update(
+                query="No relevant memory query generated, skipping search",
+            )
+            return
+
         # get memory database
         db = await Memory.get(self.agent)
 

python/extensions/message_loop_prompts_after/_91_recall_wait.py

@@ -1,19 +1,32 @@
 from python.helpers.extension import Extension
 from agent import LoopData
-from python.extensions.message_loop_prompts_after._50_recall_memories import DATA_NAME_TASK as DATA_NAME_TASK_MEMORIES
+from python.extensions.message_loop_prompts_after._50_recall_memories import DATA_NAME_TASK as DATA_NAME_TASK_MEMORIES, DATA_NAME_ITER as DATA_NAME_ITER_MEMORIES
 # from python.extensions.message_loop_prompts_after._51_recall_solutions import DATA_NAME_TASK as DATA_NAME_TASK_SOLUTIONS
-
+from python.helpers import settings
 
 class RecallWait(Extension):
     async def execute(self, loop_data: LoopData = LoopData(), **kwargs):
 
-            task = self.agent.get_data(DATA_NAME_TASK_MEMORIES)
-            if task and not task.done():
-                # self.agent.context.log.set_progress("Recalling memories...")
-                await task
+        set = settings.get_settings()
+
+        task = self.agent.get_data(DATA_NAME_TASK_MEMORIES)
+        iter = self.agent.get_data(DATA_NAME_ITER_MEMORIES) or 0
+
+        if task and not task.done():
+
+            # if memory recall is set to delayed mode, do not await on the iteration it was called
+            if set["memory_recall_delayed"]:
+                if iter == loop_data.iteration:
+                    # insert info about delayed memory to extras
+                    delay_text = self.agent.read_prompt("memory.recall_delay_msg.md")
+                    loop_data.extras_temporary["memory_recall_delayed"] = delay_text
+                    return
+            
+            # otherwise await the task
+            await task
 
-            # task = self.agent.get_data(DATA_NAME_TASK_SOLUTIONS)
-            # if task and not task.done():
-            #     # self.agent.context.log.set_progress("Recalling solutions...")
-            #     await task
+        # task = self.agent.get_data(DATA_NAME_TASK_SOLUTIONS)
+        # if task and not task.done():
+        #     # self.agent.context.log.set_progress("Recalling solutions...")
+        #     await task
 

python/extensions/monologue_start/_10_memory_init.py

@@ -0,0 +1,13 @@
+from python.helpers.extension import Extension
+from agent import LoopData
+from python.helpers import memory
+import asyncio
+
+
+class MemoryInit(Extension):
+
+    async def execute(self, loop_data: LoopData = LoopData(), **kwargs):
+        db = await memory.Memory.get(self.agent)
+        
+
+   

python/helpers/settings.py

@@ -58,6 +58,7 @@ class Settings(TypedDict):
     agent_knowledge_subdir: str
 
     memory_recall_enabled: bool
+    memory_recall_delayed: bool
     memory_recall_interval: int
     memory_recall_history_len: int
     memory_recall_memories_max_search: int
@@ -643,6 +644,16 @@ def convert_out(settings: Settings) -> SettingsOutput:
         }
     )
 
+    memory_fields.append(
+        {
+            "id": "memory_recall_delayed",
+            "title": "Memory auto-recall delayed",
+            "description": "The agent will not wait for auto memory recall. Memories will be delivered one message later. This speeds up agent's response time but may result in less relevant first step.",
+            "type": "switch",
+            "value": settings["memory_recall_delayed"],
+        }
+    )
+
     memory_fields.append(
         {
             "id": "memory_recall_query_prep",
@@ -1319,7 +1330,7 @@ def get_default_settings() -> Settings:
         chat_model_rl_input=0,
         chat_model_rl_output=0,
         util_model_provider="openrouter",
-        util_model_name="openai/gpt-5-mini",
+        util_model_name="google/gemini-2.5-flash-lite",
         util_model_api_base="",
         util_model_ctx_length=100000,
         util_model_ctx_input=0.7,
@@ -1342,6 +1353,7 @@ def get_default_settings() -> Settings:
         browser_model_rl_output=0,
         browser_model_kwargs={"temperature": "0"},
         memory_recall_enabled=True,
+        memory_recall_delayed=False,
         memory_recall_interval=3,
         memory_recall_history_len=10000,
         memory_recall_memories_max_search=12,