monocle-apptrace 0.5.3__tar.gz → 0.6.0__tar.gz

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/.gitignore

@@ -14,4 +14,9 @@ dist
 .DS_Store
 Pipfile**
 launch.json
-monocle_trace**.json
+monocle_trace**.json
+
+# Ignore LICENSE files in specific subdirectories
+apptrace/LICENSE
+mcp/LICENSE
+test_tools/LICENSE

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/PKG-INFO

@@ -1,31 +1,29 @@
 Metadata-Version: 2.4
 Name: monocle_apptrace
-Version: 0.5.3
+Version: 0.6.0
 Summary: package with monocle genAI tracing
 Project-URL: Homepage, https://github.com/monocle2ai/monocle
 Project-URL: Issues, https://github.com/monocle2ai/monocle/issues
 Author-email: "Okahu Inc." <okahu-pypi@okahu.ai>
 License: Apache-2.0
-License-File: LICENSE
-License-File: NOTICE
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
 Requires-Python: >=3.8
-Requires-Dist: click==8.2.1
-Requires-Dist: mcp>=1.13.1
 Requires-Dist: opentelemetry-api>=1.21.0
 Requires-Dist: opentelemetry-instrumentation
 Requires-Dist: opentelemetry-sdk>=1.21.0
-Requires-Dist: pydantic>=2.11.7
 Requires-Dist: requests
 Requires-Dist: wrapt>=1.14.0
+Provides-Extra: ai-test
+Requires-Dist: bert-score; extra == 'ai-test'
+Requires-Dist: transformers; extra == 'ai-test'
 Provides-Extra: aws
 Requires-Dist: boto3==1.37.24; extra == 'aws'
 Provides-Extra: azure
 Requires-Dist: azure-storage-blob==12.22.0; extra == 'azure'
 Provides-Extra: dev
-Requires-Dist: a2a-sdk==0.2.8; extra == 'dev'
+Requires-Dist: a2a-sdk==0.3.6; extra == 'dev'
 Requires-Dist: anthropic-haystack; extra == 'dev'
 Requires-Dist: anthropic==0.57.1; extra == 'dev'
 Requires-Dist: azure-storage-blob==12.22.0; extra == 'dev'
@@ -40,12 +38,13 @@ Requires-Dist: google-adk==1.10.0; extra == 'dev'
 Requires-Dist: google-generativeai==0.8.5; extra == 'dev'
 Requires-Dist: haystack-ai==2.3.0; extra == 'dev'
 Requires-Dist: httpx==0.28.1; extra == 'dev'
+Requires-Dist: huggingface-hub==0.35.3; extra == 'dev'
 Requires-Dist: instructorembedding==1.0.1; extra == 'dev'
 Requires-Dist: langchain-anthropic==0.3.13; extra == 'dev'
 Requires-Dist: langchain-aws==0.2.23; extra == 'dev'
 Requires-Dist: langchain-chroma==0.2.4; extra == 'dev'
 Requires-Dist: langchain-community==0.3.24; extra == 'dev'
-Requires-Dist: langchain-google-genai==2.1.8; extra == 'dev'
+Requires-Dist: langchain-google-genai==2.0.10; extra == 'dev'
 Requires-Dist: langchain-mcp-adapters==0.1.8; extra == 'dev'
 Requires-Dist: langchain-mistralai==0.2.10; extra == 'dev'
 Requires-Dist: langchain-openai==0.3.18; extra == 'dev'
@@ -64,6 +63,7 @@ Requires-Dist: llama-index-vector-stores-opensearch==0.6.0; extra == 'dev'
 Requires-Dist: llama-index==0.13.0; extra == 'dev'
 Requires-Dist: mcp==1.12.1; extra == 'dev'
 Requires-Dist: mistral-haystack==0.0.2; extra == 'dev'
+Requires-Dist: mistralai==1.9.9; extra == 'dev'
 Requires-Dist: numpy==1.26.4; extra == 'dev'
 Requires-Dist: openai-agents==0.2.6; extra == 'dev'
 Requires-Dist: opendal==0.45.14; extra == 'dev'
@@ -80,42 +80,12 @@ Requires-Dist: types-requests==2.31.0.20240106; extra == 'dev'
 Requires-Dist: uvicorn==0.35.0; extra == 'dev'
 Description-Content-Type: text/markdown
 
-# Monocle for tracing GenAI app code
+# Monocle Apptrace
 
 **Monocle** helps developers and platform engineers building or managing GenAI apps monitor these in prod by making it easy to instrument their code to capture traces that are compliant with open-source cloud-native observability ecosystem. 
 
 **Monocle** is a community-driven OSS framework for tracing GenAI app code governed as a [Linux Foundation AI & Data project](https://lfaidata.foundation/projects/monocle/). 
 
-## Why Monocle
-
-Monocle is built for: 
-- **app developers** to trace their app code in any environment without lots of custom code decoration 
-- **platform engineers** to instrument apps in prod through wrapping instead of asking app devs to recode
-- **GenAI component providers** to add observability features to their products 
-- **enterprises** to consume traces from GenAI apps in their existing open-source observability stack
-
-Benefits:
-- Monocle provides an implementation + package, not just a spec 
-   - No expertise in OpenTelemetry spec required
-   - No bespoke implementation of that spec required
-   - No last-mile GenAI domain specific code required to instrument your app
-- Monocle provides consistency  
-   - Connect traces across app code executions, model inference or data retrievals
-   - No cleansing of telemetry data across GenAI component providers required
-   - Works the same in personal lab dev or org cloud prod environments
-   - Send traces to location that fits your scale, budget and observability stack
-- Monocle is fully open source and community driven
-   - No vendor lock-in
-   - Implementation is transparent
-   - You can freely use or customize it to fit your needs 
-
-## What Monocle provides
-
-- Easy to [use](#use-monocle) code instrumentation
-- OpenTelemetry compatible format for [spans](src/monocle_apptrace/metamodel/spans/span_format.json). 
-- Community-curated and extensible [metamodel](src/monocle_apptrace/metamodel/README.md) for consisent tracing of GenAI components. 
-- Export to local and cloud storage 
-
 ## Use Monocle
 
 - Get the Monocle package
@@ -137,42 +107,4 @@ Benefits:
 See [Monocle user guide](Monocle_User_Guide.md) for more details.
   
 
-## Use Monocle MCP
-
-First install monocle-apptrace: pip install monocle-apptrace
-
-Open bash and run the following command to run the monocle mcp server with stdio:
-monocle_apptrace
-
-If you are using VS Code you can add following entry to your .vscode/mcp.json
-
-```json
-"monocle-mcp-server": {
-      "type": "stdio",
-      "command": "uvx",
-      "args": [
-         "monocle_apptrace"
-      ],
-      "env": {}
-   }
-```
-
-## Roadmap 
-
-Goal of Monocle is to support tracing for apps written in *any language* with *any LLM orchestration or agentic framework* and built using models, vectors, agents or other components served up by *any cloud or model inference provider*. 
-
-Current version supports: 
-- Language: (🟢) Python , (🔜) [Typescript](https://github.com/monocle2ai/monocle-typescript) 
-- LLM-frameworks: (🟢) Langchain, (🟢) Llamaindex, (🟢) Haystack, (🔜) Flask
-- LLM inference providers: (🟢) OpenAI, (🟢) Azure OpenAI, (🟢) Nvidia Triton, (🔜) AWS Bedrock, (🔜) Google Vertex, (🔜) Azure ML, (🔜) Hugging Face
-- Vector stores: (🟢) FAISS, (🔜) OpenSearch, (🔜) Milvus
-- Exporter: (🟢) stdout, (🟢) file, (🔜) Azure Blob Storage, (🔜) AWS S3, (🔜) Google Cloud Storage
-
-
-## Get involved
-### Provide feedback
-- Submit issues and enhancements requests via Github issues
-
-### Contribute
-- Monocle is community based open source project. We welcome your contributions. Please refer to the CONTRIBUTING and CODE_OF_CONDUCT for guidelines. The [contributor's guide](CONTRIBUTING.md) provides technical details of the project.
 

monocle_apptrace-0.6.0/README.md

@@ -0,0 +1,28 @@
+# Monocle Apptrace
+
+**Monocle** helps developers and platform engineers building or managing GenAI apps monitor these in prod by making it easy to instrument their code to capture traces that are compliant with open-source cloud-native observability ecosystem. 
+
+**Monocle** is a community-driven OSS framework for tracing GenAI app code governed as a [Linux Foundation AI & Data project](https://lfaidata.foundation/projects/monocle/). 
+
+## Use Monocle
+
+- Get the Monocle package
+  
+```
+    pip install monocle_apptrace 
+```
+- Instrument your app code
+     - Import the Monocle package
+       ```
+          from monocle_apptrace.instrumentor import setup_monocle_telemetry
+       ```
+     - Setup instrumentation in your ```main()``` function  
+       ``` 
+          setup_monocle_telemetry(workflow_name="your-app-name")
+       ```         
+- (Optionally) Modify config to alter where traces are sent
+
+See [Monocle user guide](Monocle_User_Guide.md) for more details.
+  
+
+

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/pyproject.toml

@@ -1,10 +1,16 @@
 [build-system]
-requires = ["hatchling"]
+requires = ["hatchling","hatch-build-scripts"]
 build-backend = "hatchling.build"
 
+[[tool.hatch.build.hooks.build-scripts.scripts]]
+commands = [
+    "echo 'Running a pre-build step..copying license'",
+    "if [ -f \"LICENSE\" ]; then echo \"✅ LICENSE already exists\"; elif [ -f \"../LICENSE\" ]; then cp \"../LICENSE\" \"LICENSE\" && echo \"✅ LICENSE copied from parent\"; else echo \"❌ ERROR: LICENSE not found\" && exit 1; fi"
+]
+
 [project]
 name = "monocle_apptrace"
-version = "0.5.3"
+version = "0.6.0"
 authors = [
   { name="Okahu Inc.", email="okahu-pypi@okahu.ai" },
 ]
@@ -23,9 +29,6 @@ dependencies = [
    'opentelemetry-api>=1.21.0',
    'opentelemetry-sdk>=1.21.0',
    'opentelemetry-instrumentation',
-   'mcp>=1.13.1',
-   'pydantic>=2.11.7',
-   'click==8.2.1'
 ]
 
 [tool.pytest.ini_options]
@@ -46,6 +49,11 @@ markers = [
 ]
 
 [project.optional-dependencies]
+ai_test = [
+   'bert-score',
+   'transformers'
+]
+
 dev = [
   'langchain-openai==0.3.18',
   'langchain-chroma==0.2.4',
@@ -89,17 +97,19 @@ dev = [
   'langchain-anthropic==0.3.13',
   'langchain-mcp-adapters==0.1.8',
   'mcp==1.12.1',
-  'a2a-sdk==0.2.8', 
+  'a2a-sdk==0.3.6',
   'httpx==0.28.1',
   'fastapi>=0.115.0',
   'pydantic==2.11.7',
   'uvicorn==0.35.0',
   'click==8.2.1',
   'python-dotenv>=1.1.0',
-  'langchain-google-genai==2.1.8',
+  'langchain-google-genai==2.0.10',
   'google-adk==1.10.0',
   'google-generativeai==0.8.5',
-  'openai-agents==0.2.6'
+  'openai-agents==0.2.6',
+  'mistralai==1.9.9',
+  'huggingface_hub==0.35.3'
 ]
 
 azure = [

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/exporters/file_exporter.py

@@ -55,6 +55,9 @@ class FileSpanExporter(SpanExporterBase):
         else:
             return self._process_spans(spans, is_root_span=is_root_span)
 
+    def set_service_name(self, service_name: str) -> None:
+        self.service_name = service_name
+
     def _cleanup_expired_handles(self) -> None:
         """Close and remove file handles that have exceeded the timeout."""
         current_time = datetime.now()
@@ -130,7 +133,10 @@ class FileSpanExporter(SpanExporterBase):
         
         # Process spans for each trace
         for trace_id, trace_spans in spans_by_trace.items():
-            service_name = trace_spans[0].resource.attributes.get(SERVICE_NAME, "unknown")
+            if self.service_name is not None:
+                service_name = self.service_name
+            else:
+                service_name = trace_spans[0].resource.attributes.get(SERVICE_NAME, "unknown")
             handle, file_path, is_first_span = self._get_or_create_handle(trace_id, service_name)
             
             if handle is None:

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/common/instrumentor.py

@@ -165,7 +165,7 @@ def setup_monocle_telemetry(
         span_handlers: Dict[str,SpanHandler] = None,
         wrapper_methods: List[Union[dict,WrapperMethod]] = None,
         union_with_default_methods: bool = True,
-        monocle_exporters_list:str = None) -> None:
+        monocle_exporters_list:str = None) -> MonocleInstrumentor:
     """
     Set up Monocle telemetry for the application.
 

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/common/span_handler.py

@@ -26,7 +26,8 @@ WORKFLOW_TYPE_MAP = {
     "anthropic": "workflow.anthropic",
     "gemini": "workflow.gemini",
     "litellm": "workflow.litellm",
-    "mistralai": "workflow.mistral"
+    "mistralai": "workflow.mistral",
+    "huggingface_hub": "workflow.huggingface"
 }
 
 FRAMEWORK_WORKFLOW_LIST = [

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/common/wrapper_method.py

@@ -5,6 +5,7 @@ from monocle_apptrace.instrumentation.common.span_handler import SpanHandler, No
 from monocle_apptrace.instrumentation.metamodel.azureaiinference.methods import AZURE_AI_INFERENCE_METHODS
 from monocle_apptrace.instrumentation.metamodel.botocore.methods import BOTOCORE_METHODS
 from monocle_apptrace.instrumentation.metamodel.botocore.handlers.botocore_span_handler import BotoCoreSpanHandler
+from monocle_apptrace.instrumentation.metamodel.hugging_face.methods import HUGGING_FACE_METHODS
 from monocle_apptrace.instrumentation.metamodel.langchain.methods import (
     LANGCHAIN_METHODS,
 )
@@ -109,7 +110,8 @@ DEFAULT_METHODS_LIST = (
     A2A_CLIENT_METHODS +
     LITELLM_METHODS +
     ADK_METHODS +
-    MISTRAL_METHODS
+    MISTRAL_METHODS +
+    HUGGING_FACE_METHODS
 )
 
 MONOCLE_SPAN_HANDLERS: Dict[str, SpanHandler] = {

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/metamodel/adk/_helper.py

@@ -82,7 +82,7 @@ def extract_agent_input(arguments: Dict[str, Any]) -> Any:
     Returns:
         Any: The extracted input data
     """
-    return arguments['args'][0].user_content.parts[0].text
+    return [arguments['args'][0].user_content.parts[0].text]
 
 def extract_agent_request_input(arguments: Dict[str, Any]) -> Any:
     """
@@ -94,7 +94,7 @@ def extract_agent_request_input(arguments: Dict[str, Any]) -> Any:
     Returns:
         Any: The extracted input data
     """
-    return arguments['kwargs']['new_message'].parts[0].text if 'new_message' in arguments['kwargs'] else None
+    return [arguments['kwargs']['new_message'].parts[0].text] if 'new_message' in arguments['kwargs'] else []
 
 def extract_agent_response(result: Any) -> Any:
     """
@@ -107,7 +107,9 @@ def extract_agent_response(result: Any) -> Any:
         Any: The extracted response data
     """
     if result:
-        return result.content.parts[0].text
+        return str(result.content.parts[0].text)
+    else:
+        return ""
 
 def get_tool_name(instance: Any) -> str:
     """
@@ -179,7 +181,7 @@ def extract_tool_input(arguments: Dict[str, Any]) -> Any:
     Returns:
         Any: The extracted input data
     """
-    return str(arguments['kwargs'].get('args'))
+    return [str(arguments['kwargs'].get('args'))]
 
 def extract_tool_response(result: Any) -> Any:
     """

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/metamodel/adk/entities/agent.py

@@ -1,6 +1,7 @@
 from monocle_apptrace.instrumentation.common.constants import SPAN_SUBTYPES, SPAN_TYPES
 from monocle_apptrace.instrumentation.common.utils import get_error_message
 from monocle_apptrace.instrumentation.metamodel.adk import _helper
+from monocle_apptrace.instrumentation.common.utils import get_error_message
 AGENT = {
       "type": SPAN_TYPES.AGENTIC_INVOCATION,
       "subtype": SPAN_SUBTYPES.ROUTING,
@@ -34,7 +35,7 @@ AGENT = {
           "attributes": [
             {
                 "_comment": "this is Agent input",
-                "attribute": "query",
+                "attribute": "input",
                 "accessor": lambda arguments: _helper.extract_agent_input(arguments)
             }
           ]
@@ -42,6 +43,10 @@ AGENT = {
         {
           "name":"data.output",
           "attributes": [
+            {
+                    "attribute": "error_code",
+                    "accessor": lambda arguments: get_error_message(arguments)
+            },
             {
                 "_comment": "this is response from LLM",
                 "attribute": "response",

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/metamodel/agents/_helper.py

@@ -50,21 +50,21 @@ def extract_agent_input(arguments):
         if len(arguments["args"]) > 1:
             input_data = arguments["args"][1]
             if isinstance(input_data, str):
-                return input_data
+                return [input_data]
             elif isinstance(input_data, list):
                 # Handle list of input items
-                return get_json_dumps(input_data)
+                return input_data
 
         # Fallback to kwargs
         if "original_input" in arguments["kwargs"]:
             input_data = arguments["kwargs"]["original_input"]
             if isinstance(input_data, str):
-                return input_data
+                return [input_data]
             elif isinstance(input_data, list):
-                return get_json_dumps(input_data)
+                return input_data
     except Exception as e:
         logger.warning("Warning: Error occurred in extract_agent_input: %s", str(e))
-    return None
+    return []
 
 
 def get_agent_name(arguments) -> str:

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/metamodel/agents/entities/inference.py

@@ -1,6 +1,7 @@
 from monocle_apptrace.instrumentation.common.constants import AGENT_REQUEST_SPAN_NAME, SPAN_SUBTYPES, SPAN_TYPES
 from monocle_apptrace.instrumentation.common.utils import get_error_message
 from monocle_apptrace.instrumentation.metamodel.agents import _helper
+from monocle_apptrace.instrumentation.common.utils import get_error_message
 
 AGENT = {
     "type": SPAN_TYPES.AGENTIC_INVOCATION,
@@ -35,7 +36,7 @@ AGENT = {
             "attributes": [
                 {
                     "_comment": "this is Agent input",
-                    "attribute": "query",
+                    "attribute": "input",
                     "accessor": lambda arguments: _helper.extract_agent_input(
                         arguments
                     ),
@@ -45,6 +46,10 @@ AGENT = {
         {
             "name": "data.output",
             "attributes": [
+                {
+                    "attribute": "error_code",
+                    "accessor": lambda arguments: get_error_message(arguments)
+                },
                 {
                     "_comment": "this is response from Agent",
                     "attribute": "response",
@@ -73,7 +78,7 @@ AGENT = {
 }
 
 AGENT_REQUEST = {
-    "type": AGENT_REQUEST_SPAN_NAME,
+    "type": SPAN_TYPES.AGENTIC_REQUEST,
     "subtype": SPAN_SUBTYPES.PLANNING,
     "attributes": [
         [

{monocle_apptrace-0.5.3 → monocle_apptrace-0.6.0}/src/monocle_apptrace/instrumentation/metamodel/finish_types.py

@@ -14,6 +14,7 @@ class FinishType(Enum):
     TOOL_CALL_ERROR = "tool_call_error"
     REFUSAL = "refusal"
     RATE_LIMITED = "rate_limited"
+    TOOL_CALL = "tool_call"
 
 # OpenAI finish reason mapping
 OPENAI_FINISH_REASON_MAPPING = {
@@ -274,6 +275,24 @@ HAYSTACK_FINISH_REASON_MAPPING = {
     "OTHER": FinishType.ERROR.value,
 }
 
+MISTRAL_FINISH_REASON_MAPPING = {
+    "stop": FinishType.SUCCESS.value,
+    "tool_calls": FinishType.TOOL_CALL.value,  # New category for tool calls
+    "length": FinishType.TRUNCATED.value,
+    # Mistral's API documentation does not explicitly mention other finish reasons like "content_filter" or "refusal".
+    # However, in case of an API-level error, the response itself would likely be an HTTP error rather than a
+    # successful response with a specific finish reason.
+}
+
+HUGGING_FACE_FINISH_REASON_MAPPING = {
+    "stop": FinishType.SUCCESS.value,
+    "tool_calls": FinishType.TOOL_CALL.value,  # New category for tool calls
+    "length": FinishType.TRUNCATED.value,
+    # Hugging Face's API documentation does not explicitly mention other finish reasons like "content_filter" or "refusal".
+    # However, in case of an API-level error, the response itself would likely be an HTTP error rather than a
+    # successful response with a specific finish reason.
+}
+
 ADK_FINISH_REASON_MAPPING = GEMINI_FINISH_REASON_MAPPING
 
 def map_openai_finish_reason_to_finish_type(finish_reason):
@@ -462,4 +481,16 @@ def map_adk_finish_reason_to_finish_type(finish_reason):
     """Map ADK finish_reason to standardized finish_type."""
     if not finish_reason:
         return None
-    return ADK_FINISH_REASON_MAPPING.get(finish_reason, None)
+    return ADK_FINISH_REASON_MAPPING.get(finish_reason, None)
+
+def map_mistral_finish_reason_to_finish_type(finish_reason):
+    """Map Mistral finish_reason to standardized finish_type."""
+    if not finish_reason:
+        return None
+    return MISTRAL_FINISH_REASON_MAPPING.get(finish_reason, None)
+
+def map_hf_finish_reason_to_finish_type(finish_reason):
+    """Map Hugging Face finish_reason to standardized finish_type."""
+    if not finish_reason:
+        return None
+    return HUGGING_FACE_FINISH_REASON_MAPPING.get(finish_reason, None)

monocle_apptrace-0.6.0/src/monocle_apptrace/instrumentation/metamodel/hugging_face/_helper.py

@@ -0,0 +1,138 @@
+import os
+import json
+import logging
+from opentelemetry.context import get_value
+from monocle_apptrace.instrumentation.common.constants import (
+    AGENT_PREFIX_KEY,
+    INFERENCE_AGENT_DELEGATION,
+    INFERENCE_TURN_END,
+    INFERENCE_TOOL_CALL,
+)
+from monocle_apptrace.instrumentation.common.utils import (
+    Option,
+    get_json_dumps,
+    try_option,
+)
+
+from monocle_apptrace.instrumentation.metamodel.finish_types import map_hf_finish_reason_to_finish_type
+
+logger = logging.getLogger(__name__)
+
+def update_input_span_events(kwargs):
+    input_text = ""
+    print("DEBUG kwargs:", kwargs)
+    if "inputs" in kwargs:
+        if isinstance(kwargs["inputs"], list):
+            input_text = " | ".join(str(i) for i in kwargs["inputs"])
+        else:
+            input_text = str(kwargs["inputs"])
+    elif "messages" in kwargs:
+        input_text = json.dumps(kwargs["messages"])
+    return {"input": input_text}  # always a dict with 'input'
+
+
+
+def update_output_span_events(result):
+    try:
+        if hasattr(result, "choices") and result.choices:
+            output = [c.message for c in result.choices]
+            output_str = json.dumps(output)
+            return output_str[:200] + "..." if len(output_str) > 200 else output_str
+    except Exception as e:
+        logger.warning("Error in update_output_span_events: %s", str(e))
+    return ""
+
+def extract_messages(kwargs):
+    """Extract system and user messages"""
+    try:
+        messages = []
+        if "system" in kwargs and isinstance(kwargs["system"], str):
+            messages.append({"system": kwargs["system"]})
+        if 'messages' in kwargs and kwargs['messages']:
+            for msg in kwargs['messages']:
+                if msg.get('content') and msg.get('role'):
+                    messages.append({msg['role']: msg['content']})
+        return [get_json_dumps(message) for message in messages]
+    except Exception as e:
+        logger.warning("Warning: Error occurred in extract_messages: %s", str(e))
+        return []
+
+def extract_assistant_message(arguments):
+    """
+    Extract the assistant message from a Mistral response or stream chunks.
+    Returns a JSON string like {"assistant": "<text>"}.
+    """
+    try:
+        result = arguments.get("result") if isinstance(arguments, dict) else arguments
+        if result is None:
+            return ""
+
+        # Handle full response
+        if hasattr(result, "choices") and result.choices:
+            msg_obj = result.choices[0].message
+            return get_json_dumps({msg_obj.role: msg_obj.content})
+
+        # Handle streaming: result might be a list of CompletionEvent chunks
+        if isinstance(result, list):
+            content = []
+            for chunk in result:
+                if hasattr(chunk, "data") and hasattr(chunk.data, "choices") and chunk.data.choices:
+                    choice = chunk.data.choices[0]
+                    if hasattr(choice, "delta") and hasattr(choice.delta, "content"):
+                        content.append(choice.delta.content or "")
+            return get_json_dumps({"assistant": "".join(content)})
+
+        return ""
+
+    except Exception as e:
+        logger.warning("Warning in extract_assistant_message: %s", str(e))
+        return ""
+    
+def update_span_from_llm_response(result, include_token_counts=False):
+    tokens = {
+        "completion_tokens": getattr(result.usage, "completion_tokens", 0),
+        "prompt_tokens": getattr(result.usage, "prompt_tokens", 0),
+        "total_tokens": getattr(result.usage, "total_tokens", 0),
+    } if include_token_counts else {}
+    # Add other metadata fields like finish_reason, etc.
+    return {**tokens, "inference_sub_type": "turn_end"}
+
+
+def get_exception_status_code(exc):
+    if exc is None:
+        return "success"
+    code = getattr(exc, "status_code", None)
+    if code == 401:
+        return "unauthorized"
+    elif code == 403:
+        return "forbidden"
+    elif code == 404:
+        return "not_found"
+    return "error"
+
+def map_finish_reason_to_finish_type(finish_reason):
+    """Map Hugging Face finish_reason to finish_type, similar to OpenAI mapping."""
+    return map_hf_finish_reason_to_finish_type(finish_reason)
+
+
+def agent_inference_type(result):
+    """
+    Simple agent inference type logic: if message contains AGENT_PREFIX_KEY,
+    mark as delegation; otherwise it's a normal turn_end.
+    """
+    try:
+        assistant_msg = extract_assistant_message(result)
+        if assistant_msg and AGENT_PREFIX_KEY in assistant_msg:
+            return INFERENCE_AGENT_DELEGATION
+    except Exception as e:
+        logger.warning("Error in agent_inference_type: %s", str(e))
+    return INFERENCE_TURN_END
+
+
+def extract_finish_reason(result):
+    try:
+        return getattr(result, "finish_reason", None)
+    except Exception:
+        return None
+
+