feat: Add LLM call events for improved observability (#2214)

* feat: Add LLM call events for improved observability

- Introduce new LLM call events: LLMCallStartedEvent, LLMCallCompletedEvent, and LLMCallFailedEvent
- Emit events for LLM calls and tool calls to provide better tracking and debugging
- Add event handling in the LLM class to track call lifecycle
- Update event bus to support new LLM-related events
- Add test cases to validate LLM event emissions

* feat: Add event handling for LLM call lifecycle events

- Implement event listeners for LLM call events in EventListener
- Add logging for LLM call start, completion, and failure events
- Import and register new LLM-specific event types

* less log

* refactor: Update LLM event response type to support Any

* refactor: Simplify LLM call completed event emission

Remove unnecessary LLMCallType conversion when emitting LLMCallCompletedEvent

* refactor: Update LLM event docstrings for clarity

Improve docstrings for LLM call events to more accurately describe their purpose and lifecycle

* feat: Add LLMCallFailedEvent emission for tool execution errors

Enhance error handling by emitting a specific event when tool execution fails during LLM calls

src/crewai/llm.py

@@ -21,6 +21,12 @@ from typing import (
 from dotenv import load_dotenv
 from pydantic import BaseModel
 
+from crewai.utilities.events.llm_events import (
+    LLMCallCompletedEvent,
+    LLMCallFailedEvent,
+    LLMCallStartedEvent,
+    LLMCallType,
+)
 from crewai.utilities.events.tool_usage_events import ToolExecutionErrorEvent
 
 with warnings.catch_warnings():
@@ -259,6 +265,15 @@ class LLM:
             >>> print(response)
             "The capital of France is Paris."
         """
+        crewai_event_bus.emit(
+            self,
+            event=LLMCallStartedEvent(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+            ),
+        )
         # Validate parameters before proceeding with the call.
         self._validate_call_params()
 
@@ -333,12 +348,13 @@ class LLM:
 
                 # --- 4) If no tool calls, return the text response
                 if not tool_calls or not available_functions:
+                    self._handle_emit_call_events(text_response, LLMCallType.LLM_CALL)
                     return text_response
 
                 # --- 5) Handle the tool call
                 tool_call = tool_calls[0]
                 function_name = tool_call.function.name
-                print("function_name", function_name)
+
                 if function_name in available_functions:
                     try:
                         function_args = json.loads(tool_call.function.arguments)
@@ -350,6 +366,7 @@ class LLM:
                     try:
                         # Call the actual tool function
                         result = fn(**function_args)
+                        self._handle_emit_call_events(result, LLMCallType.TOOL_CALL)
                         return result
 
                     except Exception as e:
@@ -365,6 +382,12 @@ class LLM:
                                 error=str(e),
                             ),
                         )
+                        crewai_event_bus.emit(
+                            self,
+                            event=LLMCallFailedEvent(
+                                error=f"Tool execution error: {str(e)}"
+                            ),
+                        )
                         return text_response
 
                 else:
@@ -374,12 +397,28 @@ class LLM:
                     return text_response
 
             except Exception as e:
+                crewai_event_bus.emit(
+                    self,
+                    event=LLMCallFailedEvent(error=str(e)),
+                )
                 if not LLMContextLengthExceededException(
                     str(e)
                 )._is_context_limit_error(str(e)):
                     logging.error(f"LiteLLM call failed: {str(e)}")
                 raise
 
+    def _handle_emit_call_events(self, response: Any, call_type: LLMCallType):
+        """Handle the events for the LLM call.
+
+        Args:
+            response (str): The response from the LLM call.
+            call_type (str): The type of call, either "tool_call" or "llm_call".
+        """
+        crewai_event_bus.emit(
+            self,
+            event=LLMCallCompletedEvent(response=response, call_type=call_type),
+        )
+
     def _format_messages_for_provider(
         self, messages: List[Dict[str, str]]
     ) -> List[Dict[str, str]]:

src/crewai/utilities/events/__init__.py

@@ -34,6 +34,7 @@ from .tool_usage_events import (
     ToolUsageEvent,
     ToolValidateInputErrorEvent,
 )
+from .llm_events import LLMCallCompletedEvent, LLMCallFailedEvent, LLMCallStartedEvent
 
 # events
 from .event_listener import EventListener

src/crewai/utilities/events/event_listener.py

@@ -4,6 +4,11 @@ from crewai.telemetry.telemetry import Telemetry
 from crewai.utilities import Logger
 from crewai.utilities.constants import EMITTER_COLOR
 from crewai.utilities.events.base_event_listener import BaseEventListener
+from crewai.utilities.events.llm_events import (
+    LLMCallCompletedEvent,
+    LLMCallFailedEvent,
+    LLMCallStartedEvent,
+)
 
 from .agent_events import AgentExecutionCompletedEvent, AgentExecutionStartedEvent
 from .crew_events import (
@@ -253,5 +258,28 @@ class EventListener(BaseEventListener):
                 #
             )
 
+        # ----------- LLM EVENTS -----------
+
+        @crewai_event_bus.on(LLMCallStartedEvent)
+        def on_llm_call_started(source, event: LLMCallStartedEvent):
+            self.logger.log(
+                f"🤖 LLM Call Started",
+                event.timestamp,
+            )
+
+        @crewai_event_bus.on(LLMCallCompletedEvent)
+        def on_llm_call_completed(source, event: LLMCallCompletedEvent):
+            self.logger.log(
+                f"✅ LLM Call Completed",
+                event.timestamp,
+            )
+
+        @crewai_event_bus.on(LLMCallFailedEvent)
+        def on_llm_call_failed(source, event: LLMCallFailedEvent):
+            self.logger.log(
+                f"❌ LLM Call Failed: '{event.error}'",
+                event.timestamp,
+            )
+
 
 event_listener = EventListener()

src/crewai/utilities/events/llm_events.py

@@ -0,0 +1,36 @@
+from enum import Enum
+from typing import Any, Dict, List, Optional, Union
+
+from crewai.utilities.events.base_events import CrewEvent
+
+
+class LLMCallType(Enum):
+    """Type of LLM call being made"""
+
+    TOOL_CALL = "tool_call"
+    LLM_CALL = "llm_call"
+
+
+class LLMCallStartedEvent(CrewEvent):
+    """Event emitted when a LLM call starts"""
+
+    type: str = "llm_call_started"
+    messages: Union[str, List[Dict[str, str]]]
+    tools: Optional[List[dict]] = None
+    callbacks: Optional[List[Any]] = None
+    available_functions: Optional[Dict[str, Any]] = None
+
+
+class LLMCallCompletedEvent(CrewEvent):
+    """Event emitted when a LLM call completes"""
+
+    type: str = "llm_call_completed"
+    response: Any
+    call_type: LLMCallType
+
+
+class LLMCallFailedEvent(CrewEvent):
+    """Event emitted when a LLM call fails"""
+
+    error: str
+    type: str = "llm_call_failed"

src/crewai/utilities/events/task_events.py

@@ -1,4 +1,4 @@
-from typing import Any, Optional
+from typing import Optional
 
 from crewai.tasks.task_output import TaskOutput
 from crewai.utilities.events.base_events import CrewEvent

tests/utilities/cassettes/test_llm_emits_call_failed_event.yaml

@@ -0,0 +1,103 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "user", "content": "Hello, how are you?"}], "model":
+      "gpt-4o-mini", "stop": []}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '102'
+      content-type:
+      - application/json
+      cookie:
+      - _cfuvid=IY8ppO70AMHr2skDSUsGh71zqHHdCQCZ3OvkPi26NBc-1740424913267-0.0.1.1-604800000;
+        __cf_bm=fU6K5KZoDmgcEuF8_yWAYKUO5fKHh6q5.wDPnna393g-1740424913-1.0.1.1-2iOaq3JVGWs439V0HxJee0IC9HdJm7dPkeJorD.AGw0YwkngRPM8rrTzn_7ht1BkbOauEezj.wPKcBz18gIYUg
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.61.0
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.61.0
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.8
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    content: "{\n  \"id\": \"chatcmpl-B4YLA2SrC2rwdVQ3U87G5a0P5lsLw\",\n  \"object\":
+      \"chat.completion\",\n  \"created\": 1740425016,\n  \"model\": \"gpt-4o-mini-2024-07-18\",\n
+      \ \"choices\": [\n    {\n      \"index\": 0,\n      \"message\": {\n        \"role\":
+      \"assistant\",\n        \"content\": \"Hello! I'm just a computer program, so
+      I don't have feelings, but I'm here and ready to help you. How can I assist
+      you today?\",\n        \"refusal\": null\n      },\n      \"logprobs\": null,\n
+      \     \"finish_reason\": \"stop\"\n    }\n  ],\n  \"usage\": {\n    \"prompt_tokens\":
+      13,\n    \"completion_tokens\": 30,\n    \"total_tokens\": 43,\n    \"prompt_tokens_details\":
+      {\n      \"cached_tokens\": 0,\n      \"audio_tokens\": 0\n    },\n    \"completion_tokens_details\":
+      {\n      \"reasoning_tokens\": 0,\n      \"audio_tokens\": 0,\n      \"accepted_prediction_tokens\":
+      0,\n      \"rejected_prediction_tokens\": 0\n    }\n  },\n  \"service_tier\":
+      \"default\",\n  \"system_fingerprint\": \"fp_709714d124\"\n}\n"
+    headers:
+      CF-RAY:
+      - 9171d4c0ed44236e-SJC
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Mon, 24 Feb 2025 19:23:38 GMT
+      Server:
+      - cloudflare
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '1954'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999978'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_ea2703502b8827e4297cd2a7bae9d9c8
+    http_version: HTTP/1.1
+    status_code: 200
+version: 1

tests/utilities/cassettes/test_llm_emits_call_started_event.yaml

@@ -0,0 +1,108 @@
+interactions:
+- request:
+    body: '{"messages": [{"role": "user", "content": "Hello, how are you?"}], "model":
+      "gpt-4o-mini", "stop": []}'
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '102'
+      content-type:
+      - application/json
+      cookie:
+      - _cfuvid=GefCcEtb_Gem93E4a9Hvt3Xyof1YQZVJAXBb9I6pEUs-1739398417375-0.0.1.1-604800000
+      host:
+      - api.openai.com
+      user-agent:
+      - OpenAI/Python 1.61.0
+      x-stainless-arch:
+      - arm64
+      x-stainless-async:
+      - 'false'
+      x-stainless-lang:
+      - python
+      x-stainless-os:
+      - MacOS
+      x-stainless-package-version:
+      - 1.61.0
+      x-stainless-raw-response:
+      - 'true'
+      x-stainless-retry-count:
+      - '0'
+      x-stainless-runtime:
+      - CPython
+      x-stainless-runtime-version:
+      - 3.12.8
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    content: "{\n  \"id\": \"chatcmpl-B4YJU8IWKGyBQtAyPDRd3SFI2flYR\",\n  \"object\":
+      \"chat.completion\",\n  \"created\": 1740424912,\n  \"model\": \"gpt-4o-mini-2024-07-18\",\n
+      \ \"choices\": [\n    {\n      \"index\": 0,\n      \"message\": {\n        \"role\":
+      \"assistant\",\n        \"content\": \"Hello! I'm just a computer program, so
+      I don't have feelings, but I'm here and ready to help you. How can I assist
+      you today?\",\n        \"refusal\": null\n      },\n      \"logprobs\": null,\n
+      \     \"finish_reason\": \"stop\"\n    }\n  ],\n  \"usage\": {\n    \"prompt_tokens\":
+      13,\n    \"completion_tokens\": 30,\n    \"total_tokens\": 43,\n    \"prompt_tokens_details\":
+      {\n      \"cached_tokens\": 0,\n      \"audio_tokens\": 0\n    },\n    \"completion_tokens_details\":
+      {\n      \"reasoning_tokens\": 0,\n      \"audio_tokens\": 0,\n      \"accepted_prediction_tokens\":
+      0,\n      \"rejected_prediction_tokens\": 0\n    }\n  },\n  \"service_tier\":
+      \"default\",\n  \"system_fingerprint\": \"fp_7fcd609668\"\n}\n"
+    headers:
+      CF-RAY:
+      - 9171d230d8ed7ae0-SJC
+      Connection:
+      - keep-alive
+      Content-Encoding:
+      - gzip
+      Content-Type:
+      - application/json
+      Date:
+      - Mon, 24 Feb 2025 19:21:53 GMT
+      Server:
+      - cloudflare
+      Set-Cookie:
+      - __cf_bm=fU6K5KZoDmgcEuF8_yWAYKUO5fKHh6q5.wDPnna393g-1740424913-1.0.1.1-2iOaq3JVGWs439V0HxJee0IC9HdJm7dPkeJorD.AGw0YwkngRPM8rrTzn_7ht1BkbOauEezj.wPKcBz18gIYUg;
+        path=/; expires=Mon, 24-Feb-25 19:51:53 GMT; domain=.api.openai.com; HttpOnly;
+        Secure; SameSite=None
+      - _cfuvid=IY8ppO70AMHr2skDSUsGh71zqHHdCQCZ3OvkPi26NBc-1740424913267-0.0.1.1-604800000;
+        path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
+      Transfer-Encoding:
+      - chunked
+      X-Content-Type-Options:
+      - nosniff
+      access-control-expose-headers:
+      - X-Request-ID
+      alt-svc:
+      - h3=":443"; ma=86400
+      cf-cache-status:
+      - DYNAMIC
+      openai-organization:
+      - crewai-iuxna1
+      openai-processing-ms:
+      - '993'
+      openai-version:
+      - '2020-10-01'
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      x-ratelimit-limit-requests:
+      - '30000'
+      x-ratelimit-limit-tokens:
+      - '150000000'
+      x-ratelimit-remaining-requests:
+      - '29999'
+      x-ratelimit-remaining-tokens:
+      - '149999978'
+      x-ratelimit-reset-requests:
+      - 2ms
+      x-ratelimit-reset-tokens:
+      - 0s
+      x-request-id:
+      - req_d9c4d49185e97b1797061efc1e55d811
+    http_version: HTTP/1.1
+    status_code: 200
+version: 1

tests/utilities/test_events.py

@@ -1,6 +1,5 @@
-import json
 from datetime import datetime
-from unittest.mock import MagicMock, patch
+from unittest.mock import patch
 
 import pytest
 from pydantic import Field
@@ -9,6 +8,7 @@ from crewai.agent import Agent
 from crewai.agents.crew_agent_executor import CrewAgentExecutor
 from crewai.crew import Crew
 from crewai.flow.flow import Flow, listen, start
+from crewai.llm import LLM
 from crewai.task import Task
 from crewai.tools.base_tool import BaseTool
 from crewai.tools.tool_usage import ToolUsage
@@ -31,6 +31,12 @@ from crewai.utilities.events.flow_events import (
     MethodExecutionFailedEvent,
     MethodExecutionStartedEvent,
 )
+from crewai.utilities.events.llm_events import (
+    LLMCallCompletedEvent,
+    LLMCallFailedEvent,
+    LLMCallStartedEvent,
+    LLMCallType,
+)
 from crewai.utilities.events.task_events import (
     TaskCompletedEvent,
     TaskFailedEvent,
@@ -495,3 +501,43 @@ def test_flow_emits_method_execution_failed_event():
     assert received_events[0].flow_name == "TestFlow"
     assert received_events[0].type == "method_execution_failed"
     assert received_events[0].error == error
+
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_llm_emits_call_started_event():
+    received_events = []
+
+    @crewai_event_bus.on(LLMCallStartedEvent)
+    def handle_llm_call_started(source, event):
+        received_events.append(event)
+
+    @crewai_event_bus.on(LLMCallCompletedEvent)
+    def handle_llm_call_completed(source, event):
+        received_events.append(event)
+
+    llm = LLM(model="gpt-4o-mini")
+    llm.call("Hello, how are you?")
+
+    assert len(received_events) == 2
+    assert received_events[0].type == "llm_call_started"
+    assert received_events[1].type == "llm_call_completed"
+
+
+@pytest.mark.vcr(filter_headers=["authorization"])
+def test_llm_emits_call_failed_event():
+    received_events = []
+
+    @crewai_event_bus.on(LLMCallFailedEvent)
+    def handle_llm_call_failed(source, event):
+        received_events.append(event)
+
+    error_message = "Simulated LLM call failure"
+    with patch.object(LLM, "_call_llm", side_effect=Exception(error_message)):
+        llm = LLM(model="gpt-4o-mini")
+        with pytest.raises(Exception) as exc_info:
+            llm.call("Hello, how are you?")
+
+        assert str(exc_info.value) == error_message
+        assert len(received_events) == 1
+        assert received_events[0].type == "llm_call_failed"
+        assert received_events[0].error == error_message