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: duplicate messages in message context (#2998)

Andy Li03061abfb3f9
diff --git a/letta/agents/letta_agent.py b/letta/agents/letta_agent.py
index 2374e47e7..d80dc4364 100644
--- a/letta/agents/letta_agent.py
+++ b/letta/agents/letta_agent.py
@@ -236,8 +236,11 @@ class LettaAgent(BaseAgent):
agent_step_span=agent_step_span,
is_final_step=(i == max_steps - 1),
)
- self.response_messages.extend(persisted_messages)
- new_in_context_messages.extend(persisted_messages)
+
+ # TODO (cliandy): handle message contexts with larger refactor and dedupe logic
+ new_message_idx = len(initial_messages) if initial_messages else 0
+ self.response_messages.extend(persisted_messages[new_message_idx:])
+ new_in_context_messages.extend(persisted_messages[new_message_idx:])
initial_messages = None
log_event("agent.stream_no_tokens.llm_response.processed") # [4^]
@@ -391,8 +394,10 @@ class LettaAgent(BaseAgent):
is_final_step=(i == max_steps - 1),
run_id=run_id,
)
- self.response_messages.extend(persisted_messages)
- new_in_context_messages.extend(persisted_messages)
+ new_message_idx = len(initial_messages) if initial_messages else 0
+ self.response_messages.extend(persisted_messages[new_message_idx:])
+ new_in_context_messages.extend(persisted_messages[new_message_idx:])
+
initial_messages = None
log_event("agent.step.llm_response.processed") # [4^]
@@ -595,8 +600,10 @@ class LettaAgent(BaseAgent):
agent_step_span=agent_step_span,
is_final_step=(i == max_steps - 1),
)
- self.response_messages.extend(persisted_messages)
- new_in_context_messages.extend(persisted_messages)
+ new_message_idx = len(initial_messages) if initial_messages else 0
+ self.response_messages.extend(persisted_messages[new_message_idx:])
+ new_in_context_messages.extend(persisted_messages[new_message_idx:])
+
initial_messages = None
# log total step time
diff --git a/letta/server/rest_api/app.py b/letta/server/rest_api/app.py
index 7c249087b..581aa20eb 100644
--- a/letta/server/rest_api/app.py
+++ b/letta/server/rest_api/app.py
@@ -385,9 +385,10 @@ def start_server(
address=host or "127.0.0.1", # Note granian address must be an ip address
port=port or REST_DEFAULT_PORT,
workers=settings.uvicorn_workers,
- # threads=
+ # runtime_blocking_threads=
+ # runtime_threads=
reload=reload or settings.uvicorn_reload,
- reload_ignore_patterns=["openapi_letta.json"],
+ reload_paths=["../letta/"],
reload_ignore_worker_failure=True,
reload_tick=4000, # set to 4s to prevent crashing on weird state
# log_level="info"