chore: bump version 0.7.2 (#2584)

Co-authored-by: Matthew Zhou <[email protected]>
Co-authored-by: Charles Packer <[email protected]>

Author: 3 people

letta/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.7.1"
+__version__ = "0.7.2"
 
 # import clients
 from letta.client.client import LocalClient, RESTClient, create_client

letta/helpers/composio_helpers.py

@@ -10,7 +10,7 @@ def get_composio_api_key(actor: User, logger: Optional[Logger] = None) -> Option
     api_keys = SandboxConfigManager().list_sandbox_env_vars_by_key(key="COMPOSIO_API_KEY", actor=actor)
     if not api_keys:
         if logger:
-            logger.warning(f"No API keys found for Composio. Defaulting to the environment variable...")
+            logger.debug(f"No API keys found for Composio. Defaulting to the environment variable...")
         if tool_settings.composio_api_key:
             return tool_settings.composio_api_key
         else:

letta/helpers/datetime_helpers.py

@@ -66,6 +66,15 @@ def get_utc_time() -> datetime:
     return datetime.now(timezone.utc)
 
 
+def get_utc_time_int() -> int:
+    return int(get_utc_time().timestamp())
+
+
+def timestamp_to_datetime(timestamp_seconds: int) -> datetime:
+    """Convert Unix timestamp in seconds to UTC datetime object"""
+    return datetime.fromtimestamp(timestamp_seconds, tz=timezone.utc)
+
+
 def format_datetime(dt):
     return dt.strftime("%Y-%m-%d %I:%M:%S %p %Z%z")
 

letta/jobs/llm_batch_job_polling.py

@@ -73,7 +73,8 @@ async def fetch_batch_items(server: SyncServer, batch_id: str, batch_resp_id: st
     """
     updates = []
     try:
-        async for item_result in server.anthropic_async_client.beta.messages.batches.results(batch_resp_id):
+        results = await server.anthropic_async_client.beta.messages.batches.results(batch_resp_id)
+        async for item_result in results:
             # Here, custom_id should be the agent_id
             item_status = map_anthropic_individual_batch_item_status_to_job_status(item_result)
             updates.append(ItemUpdateInfo(batch_id, item_result.custom_id, item_status, item_result))

letta/llm_api/anthropic.py

@@ -20,7 +20,7 @@
 )
 
 from letta.errors import BedrockError, BedrockPermissionError
-from letta.helpers.datetime_helpers import get_utc_time
+from letta.helpers.datetime_helpers import get_utc_time_int, timestamp_to_datetime
 from letta.llm_api.aws_bedrock import get_bedrock_client
 from letta.llm_api.helpers import add_inner_thoughts_to_functions
 from letta.local_llm.constants import INNER_THOUGHTS_KWARG, INNER_THOUGHTS_KWARG_DESCRIPTION
@@ -396,7 +396,7 @@ def convert_anthropic_response_to_chatcompletion(
     return ChatCompletionResponse(
         id=response.id,
         choices=[choice],
-        created=get_utc_time(),
+        created=get_utc_time_int(),
         model=response.model,
         usage=UsageStatistics(
             prompt_tokens=prompt_tokens,
@@ -451,7 +451,7 @@ def convert_anthropic_stream_event_to_chatcompletion(
                 'logprobs': None
             }
         ],
-        'created': datetime.datetime(2025, 1, 24, 0, 18, 55, tzinfo=TzInfo(UTC)),
+        'created': 1713216662,
         'model': 'gpt-4o-mini-2024-07-18',
         'system_fingerprint': 'fp_bd83329f63',
         'object': 'chat.completion.chunk'
@@ -613,7 +613,7 @@ def convert_anthropic_stream_event_to_chatcompletion(
     return ChatCompletionChunkResponse(
         id=message_id,
         choices=[choice],
-        created=get_utc_time(),
+        created=get_utc_time_int(),
         model=model,
         output_tokens=completion_chunk_tokens,
     )
@@ -920,7 +920,7 @@ def anthropic_chat_completions_process_stream(
     chat_completion_response = ChatCompletionResponse(
         id=dummy_message.id if create_message_id else TEMP_STREAM_RESPONSE_ID,
         choices=[],
-        created=dummy_message.created_at,
+        created=int(dummy_message.created_at.timestamp()),
         model=chat_completion_request.model,
         usage=UsageStatistics(
             prompt_tokens=prompt_tokens,
@@ -954,7 +954,11 @@ def anthropic_chat_completions_process_stream(
                     message_type = stream_interface.process_chunk(
                         chat_completion_chunk,
                         message_id=chat_completion_response.id if create_message_id else chat_completion_chunk.id,
-                        message_date=chat_completion_response.created if create_message_datetime else chat_completion_chunk.created,
+                        message_date=(
+                            timestamp_to_datetime(chat_completion_response.created)
+                            if create_message_datetime
+                            else timestamp_to_datetime(chat_completion_chunk.created)
+                        ),
                         # if extended_thinking is on, then reasoning_content will be flowing as chunks
                         # TODO handle emitting redacted reasoning content (e.g. as concat?)
                         expect_reasoning_content=extended_thinking,

letta/llm_api/anthropic_client.py

@@ -22,7 +22,7 @@
     LLMServerError,
     LLMUnprocessableEntityError,
 )
-from letta.helpers.datetime_helpers import get_utc_time
+from letta.helpers.datetime_helpers import get_utc_time_int
 from letta.llm_api.helpers import add_inner_thoughts_to_functions, unpack_all_inner_thoughts_from_kwargs
 from letta.llm_api.llm_client_base import LLMClientBase
 from letta.local_llm.constants import INNER_THOUGHTS_KWARG, INNER_THOUGHTS_KWARG_DESCRIPTION
@@ -403,7 +403,7 @@ def convert_response_to_chat_completion(
         chat_completion_response = ChatCompletionResponse(
             id=response.id,
             choices=[choice],
-            created=get_utc_time(),
+            created=get_utc_time_int(),
             model=response.model,
             usage=UsageStatistics(
                 prompt_tokens=prompt_tokens,

letta/llm_api/cohere.py

@@ -4,7 +4,7 @@
 
 import requests
 
-from letta.helpers.datetime_helpers import get_utc_time
+from letta.helpers.datetime_helpers import get_utc_time_int
 from letta.helpers.json_helpers import json_dumps
 from letta.local_llm.utils import count_tokens
 from letta.schemas.message import Message
@@ -207,7 +207,7 @@ def convert_cohere_response_to_chatcompletion(
     return ChatCompletionResponse(
         id=response_json["response_id"],
         choices=[choice],
-        created=get_utc_time(),
+        created=get_utc_time_int(),
         model=model,
         usage=UsageStatistics(
             prompt_tokens=prompt_tokens,

letta/llm_api/google_ai_client.py

@@ -6,7 +6,7 @@
 from google.genai.types import FunctionCallingConfig, FunctionCallingConfigMode, ToolConfig
 
 from letta.constants import NON_USER_MSG_PREFIX
-from letta.helpers.datetime_helpers import get_utc_time
+from letta.helpers.datetime_helpers import get_utc_time_int
 from letta.helpers.json_helpers import json_dumps
 from letta.llm_api.helpers import make_post_request
 from letta.llm_api.llm_client_base import LLMClientBase
@@ -260,7 +260,7 @@ def convert_response_to_chat_completion(
                 id=response_id,
                 choices=choices,
                 model=self.llm_config.model,  # NOTE: Google API doesn't pass back model in the response
-                created=get_utc_time(),
+                created=get_utc_time_int(),
                 usage=usage,
             )
         except KeyError as e:

letta/llm_api/google_vertex_client.py

@@ -4,7 +4,7 @@
 from google import genai
 from google.genai.types import FunctionCallingConfig, FunctionCallingConfigMode, GenerateContentResponse, ThinkingConfig, ToolConfig
 
-from letta.helpers.datetime_helpers import get_utc_time
+from letta.helpers.datetime_helpers import get_utc_time_int
 from letta.helpers.json_helpers import json_dumps
 from letta.llm_api.google_ai_client import GoogleAIClient
 from letta.local_llm.json_parser import clean_json_string_extra_backslash
@@ -234,7 +234,7 @@ def convert_response_to_chat_completion(
                 id=response_id,
                 choices=choices,
                 model=self.llm_config.model,  # NOTE: Google API doesn't pass back model in the response
-                created=get_utc_time(),
+                created=get_utc_time_int(),
                 usage=usage,
             )
         except KeyError as e:

letta/llm_api/openai.py

@@ -4,7 +4,9 @@
 import requests
 from openai import OpenAI
 
+from letta.helpers.datetime_helpers import timestamp_to_datetime
 from letta.llm_api.helpers import add_inner_thoughts_to_functions, convert_to_structured_output, make_post_request
+from letta.llm_api.openai_client import supports_parallel_tool_calling, supports_temperature_param
 from letta.local_llm.constants import INNER_THOUGHTS_KWARG, INNER_THOUGHTS_KWARG_DESCRIPTION, INNER_THOUGHTS_KWARG_DESCRIPTION_GO_FIRST
 from letta.local_llm.utils import num_tokens_from_functions, num_tokens_from_messages
 from letta.log import get_logger
@@ -135,7 +137,7 @@ def build_openai_chat_completions_request(
             tool_choice=tool_choice,
             user=str(user_id),
             max_completion_tokens=llm_config.max_tokens,
-            temperature=1.0 if llm_config.enable_reasoner else llm_config.temperature,
+            temperature=llm_config.temperature if supports_temperature_param(model) else None,
             reasoning_effort=llm_config.reasoning_effort,
         )
     else:
@@ -237,7 +239,7 @@ def openai_chat_completions_process_stream(
     chat_completion_response = ChatCompletionResponse(
         id=dummy_message.id if create_message_id else TEMP_STREAM_RESPONSE_ID,
         choices=[],
-        created=dummy_message.created_at,  # NOTE: doesn't matter since both will do get_utc_time()
+        created=int(dummy_message.created_at.timestamp()),  # NOTE: doesn't matter since both will do get_utc_time()
         model=chat_completion_request.model,
         usage=UsageStatistics(
             completion_tokens=0,
@@ -274,7 +276,11 @@ def openai_chat_completions_process_stream(
                     message_type = stream_interface.process_chunk(
                         chat_completion_chunk,
                         message_id=chat_completion_response.id if create_message_id else chat_completion_chunk.id,
-                        message_date=chat_completion_response.created if create_message_datetime else chat_completion_chunk.created,
+                        message_date=(
+                            timestamp_to_datetime(chat_completion_response.created)
+                            if create_message_datetime
+                            else timestamp_to_datetime(chat_completion_chunk.created)
+                        ),
                         expect_reasoning_content=expect_reasoning_content,
                         name=name,
                         message_index=message_idx,
@@ -489,6 +495,7 @@ def prepare_openai_payload(chat_completion_request: ChatCompletionRequest):
     #         except ValueError as e:
     #             warnings.warn(f"Failed to convert tool function to structured output, tool={tool}, error={e}")
 
-    if "o3-mini" in chat_completion_request.model or "o1" in chat_completion_request.model:
+    if not supports_parallel_tool_calling(chat_completion_request.model):
         data.pop("parallel_tool_calls", None)
+
     return data