Skip to content

Agent Zero AI framework

delayed memory recall

frdelaa648f0f4973
diff --git a/README.md b/README.md
index 52fad2e3f..f8da9d0cc 100644
--- a/README.md
+++ b/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
diff --git a/prompts/memory.memories_query.sys.md b/prompts/memory.memories_query.sys.md
index a39c287a1..8fed159aa 100644
--- a/prompts/memory.memories_query.sys.md
+++ b/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"
diff --git a/prompts/memory.memories_sum.sys.md b/prompts/memory.memories_sum.sys.md
index 3f9068d84..d8bde4df8 100644
--- a/prompts/memory.memories_sum.sys.md
+++ b/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)
\ No newline at end of file
+> 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.
+-
\ No newline at end of file
diff --git a/prompts/memory.recall_delay_msg.md b/prompts/memory.recall_delay_msg.md
new file mode 100644
index 000000000..15c68f01b
--- /dev/null
+++ b/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.
\ No newline at end of file
diff --git a/python/extensions/message_loop_prompts_after/_50_recall_memories.py b/python/extensions/message_loop_prompts_after/_50_recall_memories.py
index efcf1d0e1..6f95960b3 100644
--- a/python/extensions/message_loop_prompts_after/_50_recall_memories.py
+++ b/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)
diff --git a/python/extensions/message_loop_prompts_after/_91_recall_wait.py b/python/extensions/message_loop_prompts_after/_91_recall_wait.py
index 41bbb1e7b..6be6321c3 100644
--- a/python/extensions/message_loop_prompts_after/_91_recall_wait.py
+++ b/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
diff --git a/python/extensions/monologue_start/_10_memory_init.py b/python/extensions/monologue_start/_10_memory_init.py
new file mode 100644
index 000000000..26c94de30
--- /dev/null
+++ b/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)
+
+
+
\ No newline at end of file
diff --git a/python/helpers/settings.py b/python/helpers/settings.py
index 5011b1e94..965790377 100644
--- a/python/helpers/settings.py
+++ b/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,