fix: re-raise LLMError before wrapping with handle_llm_error (#9482)

LLMError exceptions are already properly formatted errors that should
propagate directly. Without this check, they get unnecessarily wrapped
by handle_llm_error, losing their original error information.

letta/adapters/simple_llm_request_adapter.py

@@ -1,6 +1,7 @@
 from typing import AsyncGenerator
 
 from letta.adapters.letta_llm_request_adapter import LettaLLMRequestAdapter
+from letta.errors import LLMError
 from letta.helpers.datetime_helpers import get_utc_timestamp_ns
 from letta.schemas.enums import LLMCallType
 from letta.schemas.letta_message import LettaMessage
@@ -54,6 +55,8 @@ class SimpleLLMRequestAdapter(LettaLLMRequestAdapter):
         try:
             self.response_data = await self.llm_client.request_async_with_telemetry(request_data, self.llm_config)
         except Exception as e:
+            if isinstance(e, LLMError):
+                raise
             raise self.llm_client.handle_llm_error(e, llm_config=self.llm_config)
 
         self.llm_request_finish_timestamp_ns = get_utc_timestamp_ns()

letta/adapters/simple_llm_stream_adapter.py

@@ -2,6 +2,7 @@ import json
 from typing import AsyncGenerator, List
 
 from letta.adapters.letta_llm_stream_adapter import LettaLLMStreamAdapter
+from letta.errors import LLMError
 from letta.log import get_logger
 
 logger = get_logger(__name__)
@@ -151,6 +152,8 @@ class SimpleLLMStreamAdapter(LettaLLMStreamAdapter):
                 error_msg=str(e),
                 error_type=type(e).__name__,
             )
+            if isinstance(e, LLMError):
+                raise
             raise self.llm_client.handle_llm_error(e, llm_config=self.llm_config)
 
         # Process the stream and yield chunks immediately for TTFT
@@ -169,6 +172,8 @@ class SimpleLLMStreamAdapter(LettaLLMStreamAdapter):
                 error_msg=str(e),
                 error_type=type(e).__name__,
             )
+            if isinstance(e, LLMError):
+                raise
             raise self.llm_client.handle_llm_error(e, llm_config=self.llm_config)
 
         # After streaming completes, extract the accumulated data

letta/agents/letta_agent.py

@@ -20,7 +20,7 @@ from letta.agents.helpers import (
     generate_step_id,
 )
 from letta.constants import DEFAULT_MAX_STEPS, NON_USER_MSG_PREFIX, REQUEST_HEARTBEAT_PARAM
-from letta.errors import ContextWindowExceededError
+from letta.errors import ContextWindowExceededError, LLMError
 from letta.helpers import ToolRulesSolver
 from letta.helpers.datetime_helpers import AsyncTimer, get_utc_time, get_utc_timestamp_ns, ns_to_ms
 from letta.helpers.reasoning_helper import scrub_inner_thoughts_from_messages
@@ -1546,6 +1546,8 @@ class LettaAgent(BaseAgent):
                 run_id=run_id,
                 step_id=step_id,
             )
+        elif isinstance(e, LLMError):
+            raise
         else:
             raise llm_client.handle_llm_error(e, llm_config=llm_config)