letta-server/letta/groups/sleeptime_multi_agent_v4.py

from collections.abc import AsyncGenerator
from datetime import datetime, timezone

from letta.agents.letta_agent_v3 import LettaAgentV3
from letta.constants import DEFAULT_MAX_STEPS
from letta.groups.helpers import stringify_message
from letta.otel.tracing import trace_method
from letta.schemas.agent import AgentState
from letta.schemas.enums import RunStatus
from letta.schemas.group import Group, ManagerType
from letta.schemas.letta_message import MessageType
from letta.schemas.letta_message_content import TextContent
from letta.schemas.letta_request import ClientToolSchema
from letta.schemas.letta_response import LettaResponse
from letta.schemas.letta_stop_reason import StopReasonType
from letta.schemas.message import Message, MessageCreate
from letta.schemas.provider_trace import BillingContext
from letta.schemas.run import Run, RunUpdate
from letta.schemas.user import User
from letta.services.group_manager import GroupManager
from letta.utils import safe_create_task


class SleeptimeMultiAgentV4(LettaAgentV3):
    def __init__(
        self,
        agent_state: AgentState,
        actor: User,
        group: Group,
    ):
        super().__init__(agent_state, actor)
        assert group.manager_type == ManagerType.sleeptime, f"Expected group type to be 'sleeptime', got {group.manager_type}"
        self.group = group
        self.run_ids = []

        # Additional manager classes
        self.group_manager = GroupManager()

    @trace_method
    async def step(
        self,
        input_messages: list[MessageCreate],
        max_steps: int = DEFAULT_MAX_STEPS,
        run_id: str | None = None,
        use_assistant_message: bool = True,
        include_return_message_types: list[MessageType] | None = None,
        request_start_timestamp_ns: int | None = None,
        conversation_id: str | None = None,
        client_tools: list[ClientToolSchema] | None = None,
        include_compaction_messages: bool = False,
        billing_context: "BillingContext | None" = None,
    ) -> LettaResponse:
        self.run_ids = []

        for i in range(len(input_messages)):
            input_messages[i].group_id = self.group.id

        response = await super().step(
            input_messages=input_messages,
            max_steps=max_steps,
            run_id=run_id,
            use_assistant_message=use_assistant_message,
            include_return_message_types=include_return_message_types,
            request_start_timestamp_ns=request_start_timestamp_ns,
            conversation_id=conversation_id,
            client_tools=client_tools,
            include_compaction_messages=include_compaction_messages,
            billing_context=billing_context,
        )

        run_ids = await self.run_sleeptime_agents()
        response.usage.run_ids = run_ids
        return response

    @trace_method
    async def stream(
        self,
        input_messages: list[MessageCreate],
        max_steps: int = DEFAULT_MAX_STEPS,
        stream_tokens: bool = True,
        run_id: str | None = None,
        use_assistant_message: bool = True,
        request_start_timestamp_ns: int | None = None,
        include_return_message_types: list[MessageType] | None = None,
        conversation_id: str | None = None,
        client_tools: list[ClientToolSchema] | None = None,
        include_compaction_messages: bool = False,
    ) -> AsyncGenerator[str, None]:
        self.run_ids = []

        for i in range(len(input_messages)):
            input_messages[i].group_id = self.group.id

        # Perform foreground agent step
        try:
            async for chunk in super().stream(
                input_messages=input_messages,
                max_steps=max_steps,
                stream_tokens=stream_tokens,
                run_id=run_id,
                use_assistant_message=use_assistant_message,
                include_return_message_types=include_return_message_types,
                request_start_timestamp_ns=request_start_timestamp_ns,
                conversation_id=conversation_id,
                client_tools=client_tools,
                include_compaction_messages=include_compaction_messages,
            ):
                yield chunk
        finally:
            # For some reason, stream is throwing a GeneratorExit even though it appears the that client
            # is getting the whole stream. This pattern should work to ensure sleeptime agents run despite this.
            await self.run_sleeptime_agents()

    @trace_method
    async def run_sleeptime_agents(self) -> list[str]:
        # Get response messages
        last_response_messages = self.response_messages

        # Update turns counter
        turns_counter = None
        if self.group.sleeptime_agent_frequency is not None and self.group.sleeptime_agent_frequency > 0:
            turns_counter = await self.group_manager.bump_turns_counter_async(group_id=self.group.id, actor=self.actor)

        # Perform participant steps
        if self.group.sleeptime_agent_frequency is None or (
            turns_counter is not None and turns_counter % self.group.sleeptime_agent_frequency == 0
        ):
            # Skip sleeptime processing if no response messages were generated
            if not last_response_messages:
                self.logger.warning("No response messages generated, skipping sleeptime agent processing")
                return self.run_ids

            last_processed_message_id = await self.group_manager.get_last_processed_message_id_and_update_async(
                group_id=self.group.id, last_processed_message_id=last_response_messages[-1].id, actor=self.actor
            )
            for sleeptime_agent_id in self.group.agent_ids:
                try:
                    sleeptime_run_id = await self._issue_background_task(
                        sleeptime_agent_id,
                        last_response_messages,
                        last_processed_message_id,
                    )
                    self.run_ids.append(sleeptime_run_id)
                except Exception as e:
                    # Individual task failures
                    print(f"Sleeptime agent processing failed: {e!s}")
                    raise e
            return self.run_ids

    @trace_method
    async def _issue_background_task(
        self,
        sleeptime_agent_id: str,
        response_messages: list[Message],
        last_processed_message_id: str,
    ) -> str:
        run = Run(
            agent_id=sleeptime_agent_id,
            status=RunStatus.created,
            metadata={
                "run_type": "sleeptime_agent_send_message_async",  # is this right?
                "agent_id": sleeptime_agent_id,
            },
        )
        run = await self.run_manager.create_run(pydantic_run=run, actor=self.actor)

        safe_create_task(
            self._participant_agent_step(
                foreground_agent_id=self.agent_state.id,
                sleeptime_agent_id=sleeptime_agent_id,
                response_messages=response_messages,
                last_processed_message_id=last_processed_message_id,
                run_id=run.id,
            ),
            label=f"participant_agent_step_{sleeptime_agent_id}",
        )
        return run.id

    @trace_method
    async def _participant_agent_step(
        self,
        foreground_agent_id: str,
        sleeptime_agent_id: str,
        response_messages: list[Message],
        last_processed_message_id: str,
        run_id: str,
    ) -> LettaResponse:
        try:
            # Update run status
            run_update = RunUpdate(status=RunStatus.running)
            await self.run_manager.update_run_by_id_async(run_id=run_id, update=run_update, actor=self.actor)

            # Create conversation transcript
            prior_messages = []
            if self.group.sleeptime_agent_frequency:
                try:
                    prior_messages = await self.message_manager.list_messages(
                        agent_id=foreground_agent_id,
                        actor=self.actor,
                        after=last_processed_message_id,
                        before=response_messages[0].id,
                    )
                except Exception:
                    pass  # continue with just latest messages

            message_strings = [stringify_message(message) for message in prior_messages + response_messages]
            message_strings = [s for s in message_strings if s is not None]
            messages_text = "\n".join(message_strings)

            message_text = (
                "<system-reminder>\n"
                "You are a sleeptime agent - a background agent that asynchronously processes conversations after they occur.\n\n"
                "IMPORTANT: You are NOT the primary agent. You are reviewing a conversation that already happened between a primary agent and its user:\n"
                '- Messages labeled "assistant" are from the primary agent (not you)\n'
                '- Messages labeled "user" are from the primary agent\'s user\n\n'
                "Your primary role is memory management. Review the conversation and use your memory tools to update any relevant memory blocks with information worth preserving. "
                "Check your memory_persona block for any additional instructions or policies.\n"
                "</system-reminder>\n\n"
                f"Messages:\n{messages_text}"
            )

            sleeptime_agent_messages = [
                MessageCreate(
                    role="user",
                    content=[TextContent(text=message_text)],
                    id=Message.generate_id(),
                    agent_id=sleeptime_agent_id,
                    group_id=self.group.id,
                )
            ]

            # Load sleeptime agent
            sleeptime_agent_state = await self.agent_manager.get_agent_by_id_async(agent_id=sleeptime_agent_id, actor=self.actor)
            sleeptime_agent = LettaAgentV3(
                agent_state=sleeptime_agent_state,
                actor=self.actor,
            )

            # Perform sleeptime agent step
            result = await sleeptime_agent.step(
                input_messages=sleeptime_agent_messages,
                run_id=run_id,
            )

            # Update run status
            run_update = RunUpdate(
                status=RunStatus.completed,
                completed_at=datetime.now(timezone.utc).replace(tzinfo=None),
                stop_reason=result.stop_reason.stop_reason if result.stop_reason else StopReasonType.end_turn,
                metadata={
                    "result": result.model_dump(mode="json"),
                    "agent_id": sleeptime_agent_state.id,
                },
            )
            await self.run_manager.update_run_by_id_async(run_id=run_id, update=run_update, actor=self.actor)
            return result
        except Exception as e:
            run_update = RunUpdate(
                status=RunStatus.failed,
                completed_at=datetime.now(timezone.utc).replace(tzinfo=None),
                stop_reason=StopReasonType.error,
                metadata={"error": str(e)},
            )
            await self.run_manager.update_run_by_id_async(run_id=run_id, update=run_update, actor=self.actor)
            raise