Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.
chore: add tracing to new agent loop
Caren Thomasb95c68414bb8
diff --git a/letta/agents/letta_agent_v2.py b/letta/agents/letta_agent_v2.pyindex 504bd3a6b..ba8d2f11c 100644--- a/letta/agents/letta_agent_v2.py+++ b/letta/agents/letta_agent_v2.py@@ -119,6 +119,7 @@ class LettaAgentV2(BaseAgentV2):agent_id=self.agent_state.id,)+ @trace_methodasync def build_request(self, input_messages: list[MessageCreate]) -> dict:"""Build the request data for an LLM call without actually executing it.@@ -146,6 +147,7 @@ class LettaAgentV2(BaseAgentV2):return request+ @trace_methodasync def step(self,input_messages: list[MessageCreate],@@ -210,6 +212,7 @@ class LettaAgentV2(BaseAgentV2):self._request_checkpoint_finish(request_span=request_span, request_start_timestamp_ns=request_start_timestamp_ns)return LettaResponse(messages=response_letta_messages, stop_reason=self.stop_reason, usage=self.usage)+ @trace_methodasync def stream(self,input_messages: list[MessageCreate],@@ -298,6 +301,7 @@ class LettaAgentV2(BaseAgentV2):for finish_chunk in self.get_finish_chunks_for_stream(self.usage, self.stop_reason):yield f"data: {finish_chunk}\n\n"+ @trace_methodasync def _step(self,messages: list[Message],@@ -588,6 +592,7 @@ class LettaAgentV2(BaseAgentV2):return maybe_approval_request, maybe_approval_responsereturn None, None+ @trace_methodasync def _check_run_cancellation(self, run_id) -> bool:try:job = await self.job_manager.get_job_by_id_async(job_id=run_id, actor=self.actor)@@ -597,6 +602,7 @@ class LettaAgentV2(BaseAgentV2):self.logger.warning(f"Failed to check job cancellation status for job {run_id}: {e}")return False+ @trace_methodasync def _refresh_messages(self, in_context_messages: list[Message]):num_messages = await self.message_manager.size_async(agent_id=self.agent_state.id,@@ -614,6 +620,7 @@ class LettaAgentV2(BaseAgentV2):in_context_messages = scrub_inner_thoughts_from_messages(in_context_messages, self.agent_state.llm_config)return in_context_messages+ @trace_methodasync def _rebuild_memory(self,in_context_messages: list[Message],@@ -702,6 +709,7 @@ class LettaAgentV2(BaseAgentV2):else:return in_context_messages+ @trace_methodasync def _get_valid_tools(self, in_context_messages: list[Message]):tools = self.agent_state.toolsself.last_function_response = self._load_last_function_response(in_context_messages)@@ -720,6 +728,7 @@ class LettaAgentV2(BaseAgentV2):)return allowed_tools+ @trace_methoddef _load_last_function_response(self, in_context_messages: list[Message]):"""Load the last function response from message history"""for msg in reversed(in_context_messages):@@ -733,6 +742,7 @@ class LettaAgentV2(BaseAgentV2):raise ValueError(f"Invalid JSON format in message: {text_content}")return None+ @trace_methoddef _request_checkpoint_start(self, request_start_timestamp_ns: int | None) -> Span | None:if request_start_timestamp_ns is not None:request_span = tracer.start_span("time_to_first_token", start_time=request_start_timestamp_ns)@@ -742,6 +752,7 @@ class LettaAgentV2(BaseAgentV2):return request_spanreturn None+ @trace_methoddef _request_checkpoint_ttft(self, request_span: Span | None, request_start_timestamp_ns: int | None) -> Span | None:if request_span:ttft_ns = get_utc_timestamp_ns() - request_start_timestamp_ns@@ -749,6 +760,7 @@ class LettaAgentV2(BaseAgentV2):return request_spanreturn None+ @trace_methoddef _request_checkpoint_finish(self, request_span: Span | None, request_start_timestamp_ns: int | None) -> None:if request_span is not None:duration_ns = get_utc_timestamp_ns() - request_start_timestamp_ns@@ -756,6 +768,7 @@ class LettaAgentV2(BaseAgentV2):request_span.end()return None+ @trace_methoddef _step_checkpoint_start(self, step_id: str) -> Tuple[StepProgression, StepMetrics, Span]:step_start_ns = get_utc_timestamp_ns()step_metrics = StepMetrics(id=step_id, step_start_ns=step_start_ns)@@ -763,6 +776,7 @@ class LettaAgentV2(BaseAgentV2):agent_step_span.set_attributes({"step_id": step_id})return StepProgression.START, step_metrics, agent_step_span+ @trace_methoddef _step_checkpoint_llm_request_start(self, step_metrics: StepMetrics, agent_step_span: Span) -> Tuple[StepProgression, StepMetrics]:llm_request_start_ns = get_utc_timestamp_ns()step_metrics.llm_request_start_ns = llm_request_start_ns@@ -772,6 +786,7 @@ class LettaAgentV2(BaseAgentV2):)return StepProgression.START, step_metrics+ @trace_methoddef _step_checkpoint_llm_request_finish(self, step_metrics: StepMetrics, agent_step_span: Span, llm_request_finish_timestamp_ns: int) -> Tuple[StepProgression, StepMetrics]:@@ -780,6 +795,7 @@ class LettaAgentV2(BaseAgentV2):agent_step_span.add_event(name="llm_request_ms", attributes={"duration_ms": ns_to_ms(llm_request_ns)})return StepProgression.RESPONSE_RECEIVED, step_metrics+ @trace_methoddef _step_checkpoint_finish(self, step_metrics: StepMetrics, agent_step_span: Span | None, run_id: str | None) -> Tuple[StepProgression, StepMetrics]:@@ -798,6 +814,7 @@ class LettaAgentV2(BaseAgentV2):self.usage.prompt_tokens += step_usage_stats.prompt_tokensself.usage.total_tokens += step_usage_stats.total_tokens+ @trace_methodasync def _handle_ai_response(self,tool_call: ToolCall,@@ -973,6 +990,7 @@ class LettaAgentV2(BaseAgentV2):return persisted_messages, continue_stepping, stop_reason+ @trace_methoddef _decide_continuation(self,agent_state: AgentState,@@ -1145,6 +1163,7 @@ class LettaAgentV2(BaseAgentV2):)return task+ @trace_methodasync def _log_request(self,request_start_timestamp_ns: int,@@ -1170,6 +1189,7 @@ class LettaAgentV2(BaseAgentV2):if request_span:request_span.end()+ @trace_methodasync def _update_agent_last_run_metrics(self, completion_time: datetime, duration_ms: float) -> None:if not settings.track_last_agent_run:returndiff --git a/letta/groups/sleeptime_multi_agent_v3.py b/letta/groups/sleeptime_multi_agent_v3.pyindex e95310e58..f9d6ad8fe 100644--- a/letta/groups/sleeptime_multi_agent_v3.py+++ b/letta/groups/sleeptime_multi_agent_v3.py@@ -5,6 +5,7 @@ from datetime import datetime, timezonefrom letta.agents.letta_agent_v2 import LettaAgentV2from letta.constants import DEFAULT_MAX_STEPSfrom letta.groups.helpers import stringify_message+from letta.otel.tracing import trace_methodfrom letta.schemas.agent import AgentStatefrom letta.schemas.enums import JobStatusfrom letta.schemas.group import Group, ManagerType@@ -33,6 +34,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):# Additional manager classesself.group_manager = GroupManager()+ @trace_methodasync def step(self,input_messages: list[MessageCreate],@@ -61,6 +63,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):response.usage.run_ids = self.run_idsreturn response+ @trace_methodasync def stream(self,input_messages: list[MessageCreate],@@ -90,6 +93,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):await self.run_sleeptime_agents(use_assistant_message=use_assistant_message)+ @trace_methodasync def run_sleeptime_agents(self, use_assistant_message: bool = True):# Get response messageslast_response_messages = self.response_messages@@ -120,6 +124,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):print(f"Sleeptime agent processing failed: {e!s}")raise e+ @trace_methodasync def _issue_background_task(self,sleeptime_agent_id: str,@@ -149,6 +154,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):)return run.id+ @trace_methodasync def _participant_agent_step(self,foreground_agent_id: str,