rasa-pro 3.13.1a17__tar.gz → 3.13.1.dev1__tar.gz

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: rasa-pro
-Version: 3.13.1a17
+Version: 3.13.1.dev1
 Summary: State-of-the-art open-core Conversational AI framework for Enterprises that natively leverages generative AI for effortless assistant development.
 Keywords: nlp,machine-learning,machine-learning-library,bot,bots,botkit,rasa conversational-agents,conversational-ai,chatbot,chatbot-framework,bot-framework
 Author: Rasa Technologies GmbH
@@ -41,11 +41,11 @@ Requires-Dist: colorhash (>=2.0,<2.1.0)
 Requires-Dist: confluent-kafka (>=2.10.0,<3.0.0)
 Requires-Dist: cryptography (>=44.0.1)
 Requires-Dist: cvg-python-sdk (>=0.5.1,<0.6.0)
-Requires-Dist: dask (>=2024.8.0,<2024.9.0)
+Requires-Dist: dask (>=2024.7.0,<2024.8.0)
 Requires-Dist: demoji (>=1.1.0,<2.0.0)
 Requires-Dist: diskcache (>=5.6.3,<5.7.0)
 Requires-Dist: dnspython (==2.6.1)
-Requires-Dist: faiss-cpu (>=1.11.0,<1.12.0)
+Requires-Dist: faiss-cpu (>=1.7.4,<2.0.0)
 Requires-Dist: fbmessenger (>=6.0.0,<6.1.0)
 Requires-Dist: github3.py (>=3.2.0,<3.3.0) ; extra == "gh-release-notes"
 Requires-Dist: gitpython (>=3.1.41,<3.2.0) ; extra == "full"
@@ -64,8 +64,9 @@ Requires-Dist: keras (==2.14.0)
 Requires-Dist: langchain (>=0.2.17,<0.3.0)
 Requires-Dist: langchain-community (>=0.2.19,<0.3.0)
 Requires-Dist: langcodes (>=3.5.0,<4.0.0)
+Requires-Dist: langfuse (>=2.60.2,<2.61.0)
 Requires-Dist: litellm (>=1.69.0,<1.70.0)
-Requires-Dist: matplotlib (>=3.9.4,<3.10.0)
+Requires-Dist: matplotlib (>=3.7,<3.8)
 Requires-Dist: mattermostwrapper (>=2.2,<2.3)
 Requires-Dist: networkx (>=3.1,<3.2)
 Requires-Dist: numpy (>=1.26.4,<1.27.0)
@@ -110,15 +111,14 @@ Requires-Dist: safetensors (>=0.4.5,<0.5.0)
 Requires-Dist: sanic (>=22.12,<22.13)
 Requires-Dist: sanic-cors (>=2.2.0,<2.3.0)
 Requires-Dist: sanic-jwt (>=1.8.0,<2.0.0)
-Requires-Dist: sanic-openapi (>=21.12.0,<22.0.0)
 Requires-Dist: sanic-routing (>=22.8.0,<23.0.0)
-Requires-Dist: scikit-learn (>=1.6.1,<1.7.0)
+Requires-Dist: scikit-learn (>=1.5.1,<1.6.0)
 Requires-Dist: scipy (>=1.13.1,<1.14.0)
 Requires-Dist: sentencepiece[sentencepiece] (>=0.1.99,<0.2.0) ; extra == "transformers" or extra == "full"
 Requires-Dist: sentry-sdk (>=2.8.0,<3)
 Requires-Dist: setuptools (>=78.1.1,<78.2.0)
-Requires-Dist: sklearn-crfsuite (>=0.5.0,<0.6.0)
-Requires-Dist: skops (>=0.11.0,<0.12.0)
+Requires-Dist: sklearn-crfsuite (>=0.3.6,<0.4.0)
+Requires-Dist: skops (>=0.10.0,<0.11.0)
 Requires-Dist: slack-sdk (>=3.27.1,<3.28.0)
 Requires-Dist: spacy (>=3.5.4,<4.0.0) ; extra == "spacy" or extra == "full"
 Requires-Dist: structlog (>=23.1.0,<23.2.0)
@@ -136,7 +136,7 @@ Requires-Dist: tensorflow-metal (==1.1.0) ; (sys_platform == "darwin" and platfo
 Requires-Dist: tensorflow-text (==2.14.0) ; sys_platform != "win32" and platform_machine != "arm64" and platform_machine != "aarch64"
 Requires-Dist: tensorflow_hub (>=0.13.0,<0.14.0)
 Requires-Dist: terminaltables (>=3.1.10,<3.2.0)
-Requires-Dist: tiktoken (>=0.9.0,<0.10.0)
+Requires-Dist: tiktoken (>=0.7.0,<0.8.0)
 Requires-Dist: tqdm (>=4.66.2,<5.0.0)
 Requires-Dist: transformers (>=4.38.2,<4.39.0) ; extra == "transformers" or extra == "full"
 Requires-Dist: twilio (>=8.4,<8.5)

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "rasa-pro"
-version = "3.13.1a17"
+version = "3.13.1.dev1"
 description = "State-of-the-art open-core Conversational AI framework for Enterprises that natively leverages generative AI for effortless assistant development."
 authors = [ "Rasa Technologies GmbH <hi@rasa.com>",]
 maintainers = [ "Tom Bocklisch <tom@rasa.com>",]
@@ -128,7 +128,7 @@ boto3 = "~1.35.15"
 google-cloud-storage = "~2.14.0"
 azure-storage-blob = ">=12.16.0, <12.17.0"
 requests = "~2.32.3"
-matplotlib = "~3.9.4"
+matplotlib = ">=3.7,<3.8"
 attrs = ">=23.1,<23.2"
 jsonpickle = ">=3.3.0,<3.4"
 redis = ">=4.6.0, <6.0"
@@ -163,7 +163,7 @@ python-socketio = ">=5.8,<6"
 python-engineio = "~4.12.2" # skip 5.0.0 upgrade
 pydot = "~1.4"
 SQLAlchemy = ">=2.0.41,<2.1.0"
-sklearn-crfsuite = "~0.5.0"
+sklearn-crfsuite = "~0.3.6"
 psycopg2-binary = ">=2.9.10,<2.10.0"
 python-dateutil = "~2.8.2"
 protobuf = "~4.25.8"
@@ -208,8 +208,8 @@ hvac = "^1.2.1"
 rich = "^13.4.2"
 cvg-python-sdk = "^0.5.1"
 python-keycloak = "^3.12.0"
-tiktoken = "~0.9.0"
-faiss-cpu = "~1.11.0"
+tiktoken = "~0.7.0"
+faiss-cpu = "^1.7.4"
 pymilvus = ">=2.4.1,<2.4.2" # pinned because 2.4.2 introduces milvus-lite dependency that fails on windows
 qdrant-client = "~1.9.1"
 psutil = "^5.9.5"
@@ -218,16 +218,16 @@ types-protobuf = "4.25.0.20240417"
 litellm = "~1.69.0"
 openpyxl = "^3.1.5"
 diskcache = "~5.6.3"
-skops = "~0.11.0"
+skops = "~0.10.0"
 safetensors = "~0.4.5"
-dask = "~2024.8.0" # later versions drop python 3.9 support
+dask = "~2024.7.0" # later versions drop python 3.9 support
 numpy = "~1.26.4"
 scipy = "~1.13.1"
-scikit-learn = "~1.6.1"
+scikit-learn = "~1.5.1"
 demoji = "^1.1.0"
 azure-identity = "~1.19.0"
 langcodes = "^3.5.0"
-sanic-openapi = "^21.12.0"
+langfuse = "~2.60.2"
 
 [[tool.poetry.dependencies.gliner]]
 version = "~0.2.20"
@@ -376,6 +376,10 @@ convention = "google"
 
 
 
+
+
+
+
 [tool.poetry.group.dev.dependencies]
 ruff = ">=0.3.5,<0.4.0"
 docker = "~7.1.0"

rasa_pro-3.13.1.dev1/rasa/__init__.py

@@ -0,0 +1,15 @@
+import logging
+
+from rasa import version
+
+# define the version before the other imports since these need it
+__version__ = version.__version__
+
+from litellm.integrations.langfuse.langfuse import LangFuseLogger
+
+from rasa.monkey_patches import litellm_langfuse_logger_init_fixed
+
+# Monkey-patch the init method as early as possible before the class is used
+LangFuseLogger.__init__ = litellm_langfuse_logger_init_fixed  # type: ignore
+
+logging.getLogger(__name__).addHandler(logging.NullHandler())

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/cli/scaffold.py

@@ -24,21 +24,10 @@ class ProjectTemplateName(Enum):
 
     DEFAULT = "default"
     TUTORIAL = "tutorial"
-    PLAIN = "plain"
-    FINANCE = "finance"
-    TELCO = "telco"
 
     def __str__(self) -> str:
         return self.value
 
-    @classmethod
-    def get_all_values(cls) -> List[str]:
-        return [name.value for name in cls]
-
-    @classmethod
-    def supported_values(cls) -> str:
-        return ", ".join(cls.get_all_values())
-
 
 template_domain_path = defaultdict(lambda: DEFAULT_DOMAIN_PATH)
 template_domain_path[ProjectTemplateName.DEFAULT] = "domain"
@@ -170,17 +159,9 @@ def create_initial_project(
     path: Text, template: ProjectTemplateName = ProjectTemplateName.DEFAULT
 ) -> None:
     """Creates directory structure and templates for initial project."""
-    import distutils.dir_util as dir_util
-
-    # clear the cache of the copy_tree function, this avoids issues if
-    # a project directory existed before and we removed folders in it
-    # with shutil.rmtree. see
-    # https://stackoverflow.com/questions/9160227/dir-util-copy-tree-fails-after-shutil-rmtree
-    if hasattr(dir_util, "_path_created"):
-        dir_util._path_created.clear()
-    else:
-        dir_util.SkipRepeatAbsolutePaths.clear()
-    dir_util.copy_tree(scaffold_path(template), path)
+    from distutils.dir_util import copy_tree
+
+    copy_tree(scaffold_path(template), path)
 
 
 def scaffold_path(template: ProjectTemplateName) -> Text:

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/core/actions/action.py

@@ -908,12 +908,10 @@ class RemoteAction(Action):
                 draft.setdefault("buttons", [])
                 draft["buttons"].extend(buttons)
 
+            # Avoid overwriting `draft` values with empty values
             response = {k: v for k, v in response.items() if v}
-
-            response.update(draft)
-            bot_messages.append(
-                create_bot_utterance(response, tracker.current_language)
-            )
+            draft.update(response)
+            bot_messages.append(create_bot_utterance(draft, tracker.current_language))
 
         return bot_messages
 

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/core/actions/direct_custom_actions_executor.py

@@ -1,5 +1,5 @@
-import os
 from functools import lru_cache
+from importlib.util import find_spec
 from typing import (
     Any,
     ClassVar,
@@ -16,6 +16,7 @@ from rasa.core.actions.custom_action_executor import (
 )
 from rasa.shared.core.domain import Domain
 from rasa.shared.core.trackers import DialogueStateTracker, EventVerbosity
+from rasa.shared.exceptions import RasaException
 from rasa.utils.endpoints import EndpointConfig
 
 structlogger = structlog.get_logger(__name__)
@@ -62,20 +63,12 @@ class DirectCustomActionExecutor(CustomActionExecutor):
             return
 
         module_name = self.action_endpoint.actions_module
-        # TODO: remove this / change back
-        structlogger.info(
-            "action.direct_custom_action_executor.register_actions_from_a_module",
-            module_name=module_name,
-            cwd=os.getcwd(),
-            ls_dir=os.listdir(os.getcwd()),
-            ls_dir_module=os.listdir(os.getcwd() + "/" + module_name),
-        )
-        # if not find_spec(module_name):
-        #     raise RasaException(
-        #         f"You've provided the custom actions module '{module_name}' "
-        #         f"to run directly by the rasa server, however this module does "
-        #         f"not exist. Please check for typos in your `endpoints.yml` file."
-        #     )
+        if not find_spec(module_name):
+            raise RasaException(
+                f"You've provided the custom actions module '{module_name}' "
+                f"to run directly by the rasa server, however this module does "
+                f"not exist. Please check for typos in your `endpoints.yml` file."
+            )
 
         self.action_executor.register_package(module_name)
         DirectCustomActionExecutor._actions_module_registered = True

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/core/channels/studio_chat.py

@@ -42,14 +42,14 @@ if TYPE_CHECKING:
     from sanic import Sanic, Websocket  # type: ignore[attr-defined]
     from socketio import AsyncServer
 
-    from rasa.core.channels.channel import UserMessage
+    from rasa.core.channels.channel import InputChannel, UserMessage
     from rasa.shared.core.trackers import DialogueStateTracker
 
 
 structlogger = structlog.get_logger()
 
 
-def tracker_as_dump(tracker: "DialogueStateTracker") -> Dict[str, Any]:
+def tracker_as_dump(tracker: "DialogueStateTracker") -> str:
     """Create a dump of the tracker state."""
     from rasa.shared.core.trackers import get_trackers_for_conversation_sessions
 
@@ -60,9 +60,8 @@ def tracker_as_dump(tracker: "DialogueStateTracker") -> Dict[str, Any]:
     else:
         last_tracker = multiple_tracker_sessions[-1]
 
-    # TODO: this is a bug: the bridge converts this back to json, but it
-    # should be json in the first place
-    return last_tracker.current_state(EventVerbosity.AFTER_RESTART)
+    state = last_tracker.current_state(EventVerbosity.AFTER_RESTART)
+    return json.dumps(state)
 
 
 def does_need_action_prediction(tracker: "DialogueStateTracker") -> bool:
@@ -149,7 +148,6 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
         from rasa.core.agent import Agent
 
         self.agent: Optional[Agent] = None
-        self.latest_tracker_session_id = None
 
         # Initialize the SocketIO input channel
         SocketIOInput.__init__(
@@ -181,9 +179,7 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
         self._register_tracker_update_hook()
 
     @classmethod
-    def from_credentials(
-        cls, credentials: Optional[Dict[Text, Any]]
-    ) -> "StudioChatInput":
+    def from_credentials(cls, credentials: Optional[Dict[Text, Any]]) -> "InputChannel":
         """Creates a StudioChatInput channel from credentials."""
         credentials = credentials or {}
 
@@ -215,11 +211,6 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
         if not self.sio:
             structlogger.error("studio_chat.on_tracker_updated.sio_not_initialized")
             return
-
-        # we need the latest session id to use it for the llm helper to get the
-        # most recent conversation the user had with the bot.
-        self.latest_tracker_session_id = sender_id
-
         await self.sio.emit("tracker", tracker_dump, room=sender_id)
 
     async def on_message_proxy(
@@ -231,14 +222,7 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
 
         Triggers a tracker update notification after processing the message.
         """
-        try:
-            await on_new_message(message)
-        except Exception as e:
-            structlogger.exception(
-                "studio_chat.on_new_message.error",
-                error=str(e),
-                sender_id=message.sender_id,
-            )
+        await on_new_message(message)
 
         if not self.agent:
             structlogger.error("studio_chat.on_message_proxy.agent_not_initialized")
@@ -291,7 +275,7 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
     async def collect_call_parameters(
         self, channel_websocket: "Websocket"
     ) -> Optional[CallParameters]:
-        """Voice method to collect call parameters."""
+        """Voice method to collect call parameters"""
         session_id = channel_websocket.session_id
         return CallParameters(session_id, "local", "local", stream_id=session_id)
 
@@ -321,7 +305,7 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
     def create_output_channel(
         self, voice_websocket: "Websocket", tts_engine: TTSEngine
     ) -> VoiceOutputChannel:
-        """Create a voice output channel."""
+        """Create a voice output channel"""
         return StudioVoiceOutputChannel(
             voice_websocket,
             tts_engine,
@@ -448,16 +432,8 @@ class StudioChatInput(SocketIOInput, VoiceInputChannel):
                     ws.put_message(data)
                 return
 
-            try:
-                # Handle text messages
-                await self.handle_user_message(sid, data, on_new_message)
-            except Exception as e:
-                structlogger.exception(
-                    "studio_chat.sio.handle_message.error",
-                    error=str(e),
-                    sid=sid,
-                )
-                await self.sio.emit("error", str(e), room=sid)
+            # Handle text messages
+            await self.handle_user_message(sid, data, on_new_message)
 
         @self.sio.on("update_tracker", namespace=self.namespace)
         async def on_update_tracker(sid: Text, data: Dict) -> None:

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/core/policies/enterprise_search_policy.py

@@ -784,19 +784,17 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
         if not os.path.exists(docs_folder) or not os.path.isdir(docs_folder):
             error_message = (
                 f"Document source directory does not exist or is not a "
-                f"directory: '{os.path.abspath(docs_folder)}'. "
+                f"directory: '{docs_folder}'. "
                 "Please specify a valid path to the documents source directory in the "
                 "vector_store configuration."
             )
             structlogger.error(
                 "enterprise_search_policy.train.faiss.invalid_source_directory",
                 message=error_message,
-                docs_folder={os.path.abspath(docs_folder)},
-                configuration_value=docs_folder,
             )
             print_error_and_exit(error_message)
 
-        docs = glob.glob(os.path.join(docs_folder, "**/*.txt"), recursive=True)
+        docs = glob.glob(os.path.join(docs_folder, "*.txt"), recursive=True)
         if not docs or len(docs) < 1:
             error_message = (
                 f"Document source directory is empty: '{docs_folder}'. "
@@ -1135,7 +1133,8 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
         embeddings_config: Dict[Text, Any],
         log_source_method: str,
     ) -> None:
-        """Perform the health checks using resolved LLM and embeddings configurations.
+        """
+        Perform the health checks using resolved LLM and embeddings configurations.
         Resolved means the configuration is either:
         - A reference to a model group that has already been expanded into
           its corresponding configuration using the information from
@@ -1164,7 +1163,8 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
 
     @classmethod
     def get_system_default_prompt_based_on_config(cls, config: Dict[str, Any]) -> str:
-        """Resolves the default prompt template for Enterprise Search Policy based on
+        """
+        Resolves the default prompt template for Enterprise Search Policy based on
         the component's configuration.
 
         - The old prompt is selected when both citation and relevancy check are either
@@ -1195,7 +1195,8 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
         relevancy_check_enabled: bool,
         citation_enabled: bool,
     ) -> str:
-        """Returns the appropriate default prompt template based on the feature flags.
+        """
+        Returns the appropriate default prompt template based on the feature flags.
 
         The selection follows this priority:
         1. If relevancy check is enabled, return the prompt that includes both relevancy

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/core/policies/flows/flow_executor.py

@@ -740,14 +740,7 @@ def _run_action_step(
             # do not log about non-existing validation actions of collect steps
             utter_action_name = render_template_variables("{{context.utter}}", context)
             if utter_action_name not in available_actions:
-                structlogger.warning(
-                    "flow.step.run.action.unknown",
-                    action=action_name,
-                    event_info=(
-                        f"The action '{action_name}' is not defined in the domain but "
-                        f"getting triggered by the flow '{step.flow_id}'."
-                    ),
-                )
+                structlogger.warning("flow.step.run.action.unknown", action=action_name)
         return ContinueFlowWithNextStep(events=initial_events)
 
 

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/dialogue_understanding/generator/flow_retrieval.py

@@ -219,6 +219,11 @@ class FlowRetrieval(EmbeddingsHealthCheckMixin):
         if self.vector_store is not None:
             with self._model_storage.write_to(self._resource) as model_path:
                 self.vector_store.save_local(model_path)
+        else:
+            structlogger.warning(
+                "flow_retrieval.persist_vector_store.not_initialized",
+                event_info="Vector store is None, not persisted.",
+            )
 
     def _persist_config(self) -> None:
         with self._model_storage.write_to(self._resource) as path:
@@ -244,16 +249,6 @@ class FlowRetrieval(EmbeddingsHealthCheckMixin):
         )
 
         flows_to_embedd = flows.exclude_link_only_flows()
-
-        if not flows_to_embedd:
-            structlogger.debug(
-                "flow_retrieval.populate_vector_store.no_flows_to_embed",
-                event_info=(
-                    "No flows to embed in the vector store, skipping population."
-                ),
-            )
-            return
-
         embeddings = self._create_embedder(self.config)
         documents = self._generate_flow_documents(flows_to_embedd, domain)
         try:
@@ -425,6 +420,10 @@ class FlowRetrieval(EmbeddingsHealthCheckMixin):
             The top k documents with similarity scores.
         """
         if self.vector_store is None:
+            structlogger.error(
+                "flow_retrieval.query_vector_store.vector_store_not_configured",
+                event_info="Vector store is not configured",
+            )
             return []
         try:
             documents_with_scores = (

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/dialogue_understanding/generator/llm_based_command_generator.py

@@ -1,6 +1,8 @@
 from abc import ABC, abstractmethod
+from asyncio import Lock
 from functools import lru_cache
 from typing import Any, Dict, List, Optional, Set, Text, Tuple, Union
+from uuid import UUID, uuid4
 
 import structlog
 from jinja2 import Environment, Template, select_autoescape
@@ -89,6 +91,9 @@ class LLMBasedCommandGenerator(
         else:
             self.flow_retrieval = None
 
+        self.sender_id_to_session_id_mapping: Dict[str, UUID] = {}
+        self._lock = Lock()
+
     ### Abstract methods
     @staticmethod
     @abstractmethod
@@ -225,8 +230,7 @@ class LLMBasedCommandGenerator(
 
     @lru_cache
     def compile_template(self, template: str) -> Template:
-        """
-        Compile the prompt template and register custom filters.
+        """Compile the prompt template and register custom filters.
 
         Compiling the template is an expensive operation,
         so we cache the result.
@@ -327,7 +331,9 @@ class LLMBasedCommandGenerator(
 
     @measure_llm_latency
     async def invoke_llm(
-        self, prompt: Union[List[dict], List[str], str]
+        self,
+        prompt: Union[List[dict], List[str], str],
+        metadata: Optional[Dict[str, Any]] = None,
     ) -> Optional[LLMResponse]:
         """Use LLM to generate a response.
 
@@ -340,6 +346,7 @@ class LLMBasedCommandGenerator(
                 - a list of messages. Each message is a string and will be formatted
                     as a user message.
                 - a single message as a string which will be formatted as user message.
+            metadata: Optional metadata to be passed to the LLM call.
 
         Returns:
             An LLMResponse object.
@@ -351,7 +358,7 @@ class LLMBasedCommandGenerator(
             self.config.get(LLM_CONFIG_KEY), self.get_default_llm_config()
         )
         try:
-            return await llm.acompletion(prompt)
+            return await llm.acompletion(prompt, metadata)
         except Exception as e:
             # unfortunately, langchain does not wrap LLM exceptions which means
             # we have to catch all exceptions here
@@ -654,3 +661,7 @@ class LLMBasedCommandGenerator(
     def get_default_llm_config() -> Dict[str, Any]:
         """Get the default LLM config for the command generator."""
         return DEFAULT_LLM_CONFIG
+
+    async def _get_or_create_session_id(self, sender_id: str) -> UUID:
+        async with self._lock:
+            return self.sender_id_to_session_id_mapping.setdefault(sender_id, uuid4())

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/dialogue_understanding/generator/llm_command_generator.py

@@ -55,7 +55,9 @@ class LLMCommandGenerator(SingleStepLLMCommandGenerator):
         )
 
     async def invoke_llm(
-        self, prompt: Union[List[dict], List[str], str]
+        self,
+        prompt: Union[List[dict], List[str], str],
+        metadata: Optional[Dict[str, Any]] = None,
     ) -> Optional[LLMResponse]:
         try:
             return await super().invoke_llm(prompt)

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/dialogue_understanding/generator/multi_step/multi_step_llm_command_generator.py

@@ -42,6 +42,9 @@ from rasa.engine.storage.resource import Resource
 from rasa.engine.storage.storage import ModelStorage
 from rasa.shared.constants import (
     EMBEDDINGS_CONFIG_KEY,
+    LANGFUSE_CUSTOM_METADATA_DICT,
+    LANGFUSE_METADATA_SESSION_ID,
+    LANGFUSE_TAGS,
     RASA_PATTERN_CANNOT_HANDLE_NOT_SUPPORTED,
     ROUTE_TO_CALM_SLOT,
 )
@@ -493,7 +496,21 @@ class MultiStepLLMCommandGenerator(LLMBasedCommandGenerator):
             prompt=prompt,
         )
 
-        response = await self.invoke_llm(prompt)
+        if tracker:
+            session_id = str(await self._get_or_create_session_id(tracker.sender_id))
+        else:
+            session_id = "unknown"
+
+        metadata = {
+            LANGFUSE_METADATA_SESSION_ID: session_id,
+            LANGFUSE_CUSTOM_METADATA_DICT: {
+                "component": self.__class__.__name__,
+                "function": "_predict_commands_for_active_flow",
+            },
+            LANGFUSE_TAGS: [self.__class__.__name__],
+        }
+
+        response = await self.invoke_llm(prompt, metadata)
         llm_response = LLMResponse.ensure_llm_response(response)
         actions = None
         if llm_response and llm_response.choices:
@@ -548,7 +565,20 @@ class MultiStepLLMCommandGenerator(LLMBasedCommandGenerator):
             prompt=prompt,
         )
 
-        response = await self.invoke_llm(prompt)
+        if tracker:
+            session_id = str(await self._get_or_create_session_id(tracker.sender_id))
+        else:
+            session_id = "unknown"
+
+        metadata = {
+            LANGFUSE_METADATA_SESSION_ID: session_id,
+            LANGFUSE_CUSTOM_METADATA_DICT: {
+                "component": self.__class__.__name__,
+                "function": "_predict_commands_for_handling_flows",
+            },
+            LANGFUSE_TAGS: [self.__class__.__name__],
+        }
+        response = await self.invoke_llm(prompt, metadata)
         llm_response = LLMResponse.ensure_llm_response(response)
         actions = None
         if llm_response and llm_response.choices:
@@ -638,7 +668,22 @@ class MultiStepLLMCommandGenerator(LLMBasedCommandGenerator):
             prompt=prompt,
         )
 
-        response = await self.invoke_llm(prompt)
+        if tracker:
+            session_id = str(await self._get_or_create_session_id(tracker.sender_id))
+
+        else:
+            session_id = "unknown"
+
+        metadata = {
+            LANGFUSE_METADATA_SESSION_ID: session_id,
+            LANGFUSE_CUSTOM_METADATA_DICT: {
+                "component": self.__class__.__name__,
+                "function": "_predict_commands_for_newly_started_flow",
+            },
+            LANGFUSE_TAGS: [self.__class__.__name__],
+        }
+
+        response = await self.invoke_llm(prompt, metadata)
         llm_response = LLMResponse.ensure_llm_response(response)
         actions = None
         if llm_response and llm_response.choices:

{rasa_pro-3.13.1a17 → rasa_pro-3.13.1.dev1}/rasa/dialogue_understanding/generator/single_step/single_step_based_llm_command_generator.py

@@ -41,6 +41,9 @@ from rasa.engine.storage.resource import Resource
 from rasa.engine.storage.storage import ModelStorage
 from rasa.shared.constants import (
     EMBEDDINGS_CONFIG_KEY,
+    LANGFUSE_CUSTOM_METADATA_DICT,
+    LANGFUSE_METADATA_SESSION_ID,
+    LANGFUSE_TAGS,
     PROMPT_TEMPLATE_CONFIG_KEY,
     ROUTE_TO_CALM_SLOT,
 )
@@ -269,7 +272,18 @@ class SingleStepBasedLLMCommandGenerator(LLMBasedCommandGenerator, ABC):
             prompt=flow_prompt,
         )
 
-        response = await self.invoke_llm(flow_prompt)
+        if tracker:
+            session_id = str(await self._get_or_create_session_id(tracker.sender_id))
+        else:
+            session_id = "unknown"
+
+        metadata = {
+            LANGFUSE_METADATA_SESSION_ID: session_id,
+            LANGFUSE_CUSTOM_METADATA_DICT: {"component": self.__class__.__name__},
+            LANGFUSE_TAGS: [self.__class__.__name__],
+        }
+
+        response = await self.invoke_llm(flow_prompt, metadata)
         llm_response = LLMResponse.ensure_llm_response(response)
         # The check for 'None' maintains compatibility with older versions
         # of LLMCommandGenerator. In previous implementations, 'invoke_llm'