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: sleeptime billing context (#10082)

Kian Jones4e11aadc7df4
diff --git a/letta/groups/sleeptime_multi_agent_v2.py b/letta/groups/sleeptime_multi_agent_v2.py
index e842e0b7f..1492b5299 100644
--- a/letta/groups/sleeptime_multi_agent_v2.py
+++ b/letta/groups/sleeptime_multi_agent_v2.py
@@ -57,6 +57,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
self.step_manager = step_manager
self.telemetry_manager = telemetry_manager
self.current_run_id = current_run_id
+ self._billing_context: BillingContext | None = None
# Group settings
assert group.manager_type == ManagerType.sleeptime, f"Expected group manager type to be 'sleeptime', got {group.manager_type}"
self.group = group
@@ -73,6 +74,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
billing_context: "BillingContext | None" = None,
) -> LettaResponse:
run_ids = []
+ self._billing_context = billing_context
# Prepare new messages
new_messages = []
@@ -126,6 +128,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
last_response_messages,
last_processed_message_id,
use_assistant_message,
+ billing_context,
)
run_ids.append(run_id)
@@ -168,6 +171,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
use_assistant_message: bool = True,
request_start_timestamp_ns: int | None = None,
include_return_message_types: list[MessageType] | None = None,
+ billing_context: BillingContext | None = None,
) -> AsyncGenerator[str, None]:
# Prepare new messages
new_messages = []
@@ -197,6 +201,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
use_assistant_message=use_assistant_message,
request_start_timestamp_ns=request_start_timestamp_ns,
include_return_message_types=include_return_message_types,
+ billing_context=billing_context,
):
yield chunk
@@ -220,6 +225,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
last_response_messages,
last_processed_message_id,
use_assistant_message,
+ billing_context,
)
async def _issue_background_task(
@@ -228,6 +234,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
response_messages: list[Message],
last_processed_message_id: str,
use_assistant_message: bool = True,
+ billing_context: BillingContext | None = None,
) -> str:
run = Run(
user_id=self.actor.id,
@@ -247,6 +254,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
last_processed_message_id=last_processed_message_id,
run_id=run.id,
use_assistant_message=True,
+ billing_context=billing_context,
),
label=f"participant_agent_step_{sleeptime_agent_id}",
)
@@ -260,6 +268,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
last_processed_message_id: str,
run_id: str,
use_assistant_message: bool = True,
+ billing_context: BillingContext | None = None,
) -> str:
try:
# Update job status
@@ -315,6 +324,7 @@ class SleeptimeMultiAgentV2(BaseAgent):
input_messages=sleeptime_agent_messages,
use_assistant_message=use_assistant_message,
run_id=run_id,
+ billing_context=billing_context,
)
# Update job status
diff --git a/letta/groups/sleeptime_multi_agent_v3.py b/letta/groups/sleeptime_multi_agent_v3.py
index 4ddb2c08b..4be3504f8 100644
--- a/letta/groups/sleeptime_multi_agent_v3.py
+++ b/letta/groups/sleeptime_multi_agent_v3.py
@@ -36,6 +36,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
# Additional manager classes
self.group_manager = GroupManager()
+ self._billing_context: BillingContext | None = None
@trace_method
async def step(
@@ -52,6 +53,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
billing_context: "BillingContext | None" = None,
) -> LettaResponse:
self.run_ids = []
+ self._billing_context = billing_context
for i in range(len(input_messages)):
input_messages[i].group_id = self.group.id
@@ -69,7 +71,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
billing_context=billing_context,
)
- await self.run_sleeptime_agents()
+ await self.run_sleeptime_agents(billing_context=billing_context)
response.usage.run_ids = self.run_ids
return response
@@ -91,6 +93,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
openai_responses_websocket: bool = False,
) -> AsyncGenerator[str, None]:
self.run_ids = []
+ self._billing_context = billing_context
for i in range(len(input_messages)):
input_messages[i].group_id = self.group.id
@@ -114,13 +117,15 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
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()
+ await self.run_sleeptime_agents(billing_context=billing_context)
@trace_method
- async def run_sleeptime_agents(self):
+ async def run_sleeptime_agents(self, billing_context: BillingContext | None = None):
# Get response messages
last_response_messages = self.response_messages
+ billing_context = billing_context or self._billing_context
+
# Update turns counter
turns_counter = None
if self.group.sleeptime_agent_frequency is not None and self.group.sleeptime_agent_frequency > 0:
@@ -139,6 +144,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
sleeptime_agent_id,
last_response_messages,
last_processed_message_id,
+ billing_context,
)
self.run_ids.append(sleeptime_run_id)
except Exception as e:
@@ -152,6 +158,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
sleeptime_agent_id: str,
response_messages: list[Message],
last_processed_message_id: str,
+ billing_context: BillingContext | None,
) -> str:
run = Run(
agent_id=sleeptime_agent_id,
@@ -170,6 +177,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
response_messages=response_messages,
last_processed_message_id=last_processed_message_id,
run_id=run.id,
+ billing_context=billing_context,
),
label=f"participant_agent_step_{sleeptime_agent_id}",
)
@@ -183,6 +191,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
response_messages: list[Message],
last_processed_message_id: str,
run_id: str,
+ billing_context: BillingContext | None,
) -> LettaResponse:
try:
# Update run status
@@ -227,6 +236,7 @@ class SleeptimeMultiAgentV3(LettaAgentV2):
result = await sleeptime_agent.step(
input_messages=sleeptime_agent_messages,
run_id=run_id,
+ billing_context=billing_context,
)
# Update run status
diff --git a/letta/groups/sleeptime_multi_agent_v4.py b/letta/groups/sleeptime_multi_agent_v4.py
index 8ee71388b..a2f8ecc56 100644
--- a/letta/groups/sleeptime_multi_agent_v4.py
+++ b/letta/groups/sleeptime_multi_agent_v4.py
@@ -36,6 +36,10 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
# Additional manager classes
self.group_manager = GroupManager()
+ # Preserve the request billing context for follow-up sleeptime runs that are
+ # triggered entirely within core and therefore never traverse cloud-api.
+ self._billing_context: BillingContext | None = None
+
@trace_method
async def step(
self,
@@ -52,6 +56,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
billing_context: "BillingContext | None" = None,
) -> LettaResponse:
self.run_ids = []
+ self._billing_context = billing_context
for i in range(len(input_messages)):
input_messages[i].group_id = self.group.id
@@ -70,7 +75,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
billing_context=billing_context,
)
- run_ids = await self.run_sleeptime_agents()
+ run_ids = await self.run_sleeptime_agents(billing_context=billing_context)
response.usage.run_ids = run_ids
return response
@@ -92,6 +97,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
openai_responses_websocket: bool = False,
) -> AsyncGenerator[str, None]:
self.run_ids = []
+ self._billing_context = billing_context
for i in range(len(input_messages)):
input_messages[i].group_id = self.group.id
@@ -116,13 +122,15 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
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()
+ await self.run_sleeptime_agents(billing_context=billing_context)
@trace_method
- async def run_sleeptime_agents(self) -> list[str]:
+ async def run_sleeptime_agents(self, billing_context: BillingContext | None = None) -> list[str]:
# Get response messages
last_response_messages = self.response_messages
+ billing_context = billing_context or self._billing_context
+
# Update turns counter
turns_counter = None
if self.group.sleeptime_agent_frequency is not None and self.group.sleeptime_agent_frequency > 0:
@@ -146,6 +154,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
sleeptime_agent_id,
last_response_messages,
last_processed_message_id,
+ billing_context,
)
self.run_ids.append(sleeptime_run_id)
except Exception as e:
@@ -160,6 +169,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
sleeptime_agent_id: str,
response_messages: list[Message],
last_processed_message_id: str,
+ billing_context: BillingContext | None,
) -> str:
run = Run(
agent_id=sleeptime_agent_id,
@@ -178,6 +188,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
response_messages=response_messages,
last_processed_message_id=last_processed_message_id,
run_id=run.id,
+ billing_context=billing_context,
),
label=f"participant_agent_step_{sleeptime_agent_id}",
)
@@ -191,6 +202,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
response_messages: list[Message],
last_processed_message_id: str,
run_id: str,
+ billing_context: BillingContext | None,
) -> LettaResponse:
try:
# Update run status
@@ -247,6 +259,7 @@ class SleeptimeMultiAgentV4(LettaAgentV3):
result = await sleeptime_agent.step(
input_messages=sleeptime_agent_messages,
run_id=run_id,
+ billing_context=billing_context,
)
# Update run status
diff --git a/letta/server/rest_api/proxy_helpers.py b/letta/server/rest_api/proxy_helpers.py
index 644132b5e..d7eec5aba 100644
--- a/letta/server/rest_api/proxy_helpers.py
+++ b/letta/server/rest_api/proxy_helpers.py
@@ -338,6 +338,7 @@ async def persist_messages_background(
assistant_message: str,
model_name: str,
proxy_name: str,
+ billing_context=None,
):
"""
Background task to persist messages without blocking the response.
@@ -361,6 +362,7 @@ async def persist_messages_background(
user_messages=user_messages,
assistant_message=assistant_message,
model=model_name,
+ billing_context=billing_context,
)
if result.get("success"):
logger.info(f"[{proxy_name}] Persisted messages: {result['messages_created']} messages saved")
diff --git a/letta/server/rest_api/routers/v1/agents.py b/letta/server/rest_api/routers/v1/agents.py
index 88b0325f0..d0440805f 100644
--- a/letta/server/rest_api/routers/v1/agents.py
+++ b/letta/server/rest_api/routers/v1/agents.py
@@ -2536,6 +2536,6 @@ async def capture_messages(
if sleeptime_group:
sleeptime_agent_loop = SleeptimeMultiAgentV4(agent_state=agent, actor=actor, group=sleeptime_group)
sleeptime_agent_loop.response_messages = response_messages
- run_ids = await sleeptime_agent_loop.run_sleeptime_agents()
+ run_ids = await sleeptime_agent_loop.run_sleeptime_agents(billing_context=headers.billing_context)
return JSONResponse({"success": True, "messages_created": len(response_messages), "run_ids": run_ids})
diff --git a/letta/server/rest_api/utils.py b/letta/server/rest_api/utils.py
index 8d2fff69e..fd4ebf35c 100644
--- a/letta/server/rest_api/utils.py
+++ b/letta/server/rest_api/utils.py
@@ -37,6 +37,7 @@ from letta.schemas.letta_message_content import (
)
from letta.schemas.llm_config import LLMConfig
from letta.schemas.message import ApprovalCreate, Message, MessageCreate, ToolReturn
+from letta.schemas.provider_trace import BillingContext
from letta.schemas.tool_execution_result import ToolExecutionResult
from letta.schemas.usage import LettaUsageStatistics
from letta.schemas.user import User
@@ -801,6 +802,7 @@ async def capture_and_persist_messages(
user_messages: list[str],
assistant_message: str,
model: Optional[str] = None,
+ billing_context: BillingContext | None = None,
) -> Dict[str, Any]:
"""
Capture user and assistant messages and persist them to the database.
@@ -862,7 +864,7 @@ async def capture_and_persist_messages(
sleeptime_agent_loop = SleeptimeMultiAgentV4(agent_state=agent, actor=actor, group=sleeptime_group)
sleeptime_agent_loop.response_messages = response_messages
- run_ids = await sleeptime_agent_loop.run_sleeptime_agents()
+ run_ids = await sleeptime_agent_loop.run_sleeptime_agents(billing_context=billing_context)
logger.info(f"Triggered sleeptime agents, run_ids: {run_ids}")
except Exception as e: