Skip to content

armgit / letta

Public Imported from GitHub
0

Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.

fix: avoid using summarization agent when not using voice (#4097)

Sarah Wooders9eb303b5f443
diff --git a/letta/agents/letta_agent.py b/letta/agents/letta_agent.py
index 6ea255d7d..1b57b2c25 100644
--- a/letta/agents/letta_agent.py
+++ b/letta/agents/letta_agent.py
@@ -137,6 +137,10 @@ class LettaAgent(BaseAgent):
message_buffer_limit=message_buffer_limit,
message_buffer_min=message_buffer_min,
partial_evict_summarizer_percentage=partial_evict_summarizer_percentage,
+ agent_manager=self.agent_manager,
+ message_manager=self.message_manager,
+ actor=self.actor,
+ agent_id=self.agent_id,
)
async def _check_run_cancellation(self) -> bool:
diff --git a/letta/services/summarizer/summarizer.py b/letta/services/summarizer/summarizer.py
index b1d7c0aeb..4a57e6c53 100644
--- a/letta/services/summarizer/summarizer.py
+++ b/letta/services/summarizer/summarizer.py
@@ -15,6 +15,8 @@ from letta.schemas.letta_message_content import TextContent
from letta.schemas.llm_config import LLMConfig
from letta.schemas.message import Message, MessageCreate
from letta.schemas.user import User
+from letta.services.agent_manager import AgentManager
+from letta.services.message_manager import MessageManager
from letta.services.summarizer.enums import SummarizationMode
from letta.system import package_summarize_message_no_counts
from letta.templates.template_helper import render_template
@@ -36,6 +38,10 @@ class Summarizer:
message_buffer_limit: int = 10,
message_buffer_min: int = 3,
partial_evict_summarizer_percentage: float = 0.30,
+ agent_manager: Optional[AgentManager] = None,
+ message_manager: Optional[MessageManager] = None,
+ actor: Optional[User] = None,
+ agent_id: Optional[str] = None,
):
self.mode = mode
@@ -46,6 +52,12 @@ class Summarizer:
self.summarizer_agent = summarizer_agent
self.partial_evict_summarizer_percentage = partial_evict_summarizer_percentage
+ # for partial buffer only
+ self.agent_manager = agent_manager
+ self.message_manager = message_manager
+ self.actor = actor
+ self.agent_id = agent_id
+
@trace_method
async def summarize(
self,
@@ -121,9 +133,6 @@ class Summarizer:
logger.debug("Not forcing summarization, returning in-context messages as is.")
return all_in_context_messages, False
- # Very ugly code to pull LLMConfig etc from the SummarizerAgent if we're not using it for anything else
- assert self.summarizer_agent is not None
-
# First step: determine how many messages to retain
total_message_count = len(all_in_context_messages)
assert self.partial_evict_summarizer_percentage >= 0.0 and self.partial_evict_summarizer_percentage <= 1.0
@@ -147,15 +156,13 @@ class Summarizer:
# Dynamically get the LLMConfig from the summarizer agent
# Pretty cringe code here that we need the agent for this but we don't use it
- agent_state = await self.summarizer_agent.agent_manager.get_agent_by_id_async(
- agent_id=self.summarizer_agent.agent_id, actor=self.summarizer_agent.actor
- )
+ agent_state = await self.agent_manager.get_agent_by_id_async(agent_id=self.agent_id, actor=self.actor)
# TODO if we do this via the "agent", then we can more easily allow toggling on the memory block version
summary_message_str = await simple_summary(
messages=messages_to_summarize,
llm_config=agent_state.llm_config,
- actor=self.summarizer_agent.actor,
+ actor=self.actor,
include_ack=True,
)
@@ -185,9 +192,9 @@ class Summarizer:
)[0]
# Create the message in the DB
- await self.summarizer_agent.message_manager.create_many_messages_async(
+ await self.message_manager.create_many_messages_async(
pydantic_msgs=[summary_message_obj],
- actor=self.summarizer_agent.actor,
+ actor=self.actor,
)
updated_in_context_messages = all_in_context_messages[assistant_message_index:]