datarobot-genai 0.1.71__tar.gz → 0.2.0__tar.gz

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datarobot-genai
-Version: 0.1.71
+Version: 0.2.0
 Summary: Generic helpers for GenAI
 Project-URL: Homepage, https://github.com/datarobot-oss/datarobot-genai
 Author: DataRobot, Inc.
@@ -8,6 +8,7 @@ License: Apache-2.0
 License-File: AUTHORS
 License-File: LICENSE
 Requires-Python: <3.13,>=3.10
+Requires-Dist: ag-ui-protocol<0.2.0,>=0.1.9
 Requires-Dist: datarobot-drum<2.0.0,>=1.17.5
 Requires-Dist: datarobot-predict<2.0.0,>=1.13.2
 Requires-Dist: datarobot<4.0.0,>=3.9.1
@@ -23,7 +24,7 @@ Requires-Dist: ragas<0.4.0,>=0.3.8
 Requires-Dist: requests<3.0.0,>=2.32.4
 Provides-Extra: crewai
 Requires-Dist: crewai-tools[mcp]<0.77.0,>=0.69.0; extra == 'crewai'
-Requires-Dist: crewai<1.0.0,>=0.193.2; extra == 'crewai'
+Requires-Dist: crewai>=1.1.0; extra == 'crewai'
 Requires-Dist: opentelemetry-instrumentation-crewai<1.0.0,>=0.40.5; extra == 'crewai'
 Requires-Dist: pybase64<2.0.0,>=1.4.2; extra == 'crewai'
 Provides-Extra: drmcp
@@ -56,8 +57,8 @@ Requires-Dist: llama-index<0.14.0,>=0.13.6; extra == 'llamaindex'
 Requires-Dist: opentelemetry-instrumentation-llamaindex<1.0.0,>=0.40.5; extra == 'llamaindex'
 Requires-Dist: pypdf<7.0.0,>=6.0.0; extra == 'llamaindex'
 Provides-Extra: nat
+Requires-Dist: crewai>=1.1.0; (python_version >= '3.11') and extra == 'nat'
 Requires-Dist: llama-index-llms-litellm<0.7.0,>=0.4.1; extra == 'nat'
-Requires-Dist: nvidia-nat-crewai==1.3.0; (python_version >= '3.11') and extra == 'nat'
 Requires-Dist: nvidia-nat-langchain==1.3.0; (python_version >= '3.11') and extra == 'nat'
 Requires-Dist: nvidia-nat-opentelemetry==1.3.0; (python_version >= '3.11') and extra == 'nat'
 Requires-Dist: nvidia-nat==1.3.0; (python_version >= '3.11') and extra == 'nat'

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "datarobot-genai"
-version = "0.1.71"
+version = "0.2.0"
 description = "Generic helpers for GenAI"
 readme = "README.md"
 requires-python = ">=3.10, <3.13"
@@ -24,6 +24,7 @@ dependencies = [
   "opentelemetry-instrumentation-aiohttp-client>=0.43b0,<1.0.0",
   "opentelemetry-instrumentation-httpx>=0.43b0,<1.0.0",
   "opentelemetry-instrumentation-openai>=0.40.5,<1.0.0",
+  "ag-ui-protocol>=0.1.9,<0.2.0",
 ]
 
 [project.urls]
@@ -35,7 +36,7 @@ datarobot_llm_clients = "datarobot_genai.nat.datarobot_llm_clients"
 
 [project.optional-dependencies]
 crewai = [
-  "crewai>=0.193.2,<1.0.0",
+  "crewai>=1.1.0",
   "crewai-tools[mcp]>=0.69.0,<0.77.0",
   "opentelemetry-instrumentation-crewai>=0.40.5,<1.0.0",
   "pybase64>=1.4.2,<2.0.0",
@@ -59,8 +60,8 @@ llamaindex = [
 nat = [
   "nvidia-nat==1.3.0; python_version >= '3.11'",
   "nvidia-nat-opentelemetry==1.3.0; python_version >= '3.11'",
-  "nvidia-nat-crewai==1.3.0; python_version >= '3.11'",
   "nvidia-nat-langchain==1.3.0; python_version >= '3.11'",
+  "crewai>=1.1.0; python_version >= '3.11'",
   "llama-index-llms-litellm>=0.4.1,<0.7.0",  # Need this to support datarobot-llm plugin
   "opentelemetry-instrumentation-crewai>=0.40.5,<1.0.0",
   "opentelemetry-instrumentation-llamaindex>=0.40.5,<1.0.0",

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/src/datarobot_genai/core/agents/base.py

@@ -23,6 +23,7 @@ from typing import TypedDict
 from typing import TypeVar
 from typing import cast
 
+from ag_ui.core import Event
 from openai.types.chat import CompletionCreateParams
 from ragas import MultiTurnSample
 
@@ -167,7 +168,7 @@ class UsageMetrics(TypedDict):
 
 # Canonical return type for DRUM-compatible invoke implementations
 InvokeReturn = (
-    AsyncGenerator[tuple[str, MultiTurnSample | None, UsageMetrics], None]
+    AsyncGenerator[tuple[str | Event, MultiTurnSample | None, UsageMetrics], None]
     | tuple[str, MultiTurnSample | None, UsageMetrics]
 )
 

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/src/datarobot_genai/core/chat/responses.py

@@ -14,14 +14,23 @@
 
 """OpenAI-compatible response helpers for chat interactions."""
 
+import asyncio
+import queue
 import time
 import traceback as tb
 import uuid
 from asyncio import AbstractEventLoop
 from collections.abc import AsyncGenerator
+from collections.abc import AsyncIterator
 from collections.abc import Iterator
 from concurrent.futures import ThreadPoolExecutor
+from typing import Any
+from typing import TypeVar
 
+from ag_ui.core import BaseEvent
+from ag_ui.core import Event
+from ag_ui.core import TextMessageChunkEvent
+from ag_ui.core import TextMessageContentEvent
 from openai.types import CompletionUsage
 from openai.types.chat import ChatCompletion
 from openai.types.chat import ChatCompletionChunk
@@ -40,6 +49,7 @@ class CustomModelChatResponse(ChatCompletion):
 
 class CustomModelStreamingResponse(ChatCompletionChunk):
     pipeline_interactions: str | None = None
+    event: Event | None = None
 
 
 def to_custom_model_chat_response(
@@ -83,7 +93,7 @@ def to_custom_model_streaming_response(
     thread_pool_executor: ThreadPoolExecutor,
     event_loop: AbstractEventLoop,
     streaming_response_generator: AsyncGenerator[
-        tuple[str, MultiTurnSample | None, dict[str, int]], None
+        tuple[str | Event, MultiTurnSample | None, dict[str, int]], None
     ],
     model: str | object | None,
 ) -> Iterator[CustomModelStreamingResponse]:
@@ -105,7 +115,7 @@ def to_custom_model_streaming_response(
         while True:
             try:
                 (
-                    response_text,
+                    response_text_or_event,
                     pipeline_interactions,
                     usage_metrics,
                 ) = thread_pool_executor.submit(
@@ -114,10 +124,10 @@ def to_custom_model_streaming_response(
                 last_pipeline_interactions = pipeline_interactions
                 last_usage_metrics = usage_metrics
 
-                if response_text:
+                if isinstance(response_text_or_event, str) and response_text_or_event:
                     choice = ChunkChoice(
                         index=0,
-                        delta=ChoiceDelta(role="assistant", content=response_text),
+                        delta=ChoiceDelta(role="assistant", content=response_text_or_event),
                         finish_reason=None,
                     )
                     yield CustomModelStreamingResponse(
@@ -130,6 +140,29 @@ def to_custom_model_streaming_response(
                         if usage_metrics
                         else None,
                     )
+                elif isinstance(response_text_or_event, BaseEvent):
+                    content = ""
+                    if isinstance(
+                        response_text_or_event, (TextMessageContentEvent, TextMessageChunkEvent)
+                    ):
+                        content = response_text_or_event.delta or content
+                    choice = ChunkChoice(
+                        index=0,
+                        delta=ChoiceDelta(role="assistant", content=content),
+                        finish_reason=None,
+                    )
+
+                    yield CustomModelStreamingResponse(
+                        id=completion_id,
+                        object="chat.completion.chunk",
+                        created=created,
+                        model=model,
+                        choices=[choice],
+                        usage=CompletionUsage.model_validate(required_usage_metrics | usage_metrics)
+                        if usage_metrics
+                        else None,
+                        event=response_text_or_event,
+                    )
             except StopAsyncIteration:
                 break
         event_loop.run_until_complete(streaming_response_generator.aclose())
@@ -168,3 +201,97 @@ def to_custom_model_streaming_response(
             choices=[choice],
             usage=None,
         )
+
+
+def streaming_iterator_to_custom_model_streaming_response(
+    streaming_response_iterator: Iterator[tuple[str, MultiTurnSample | None, dict[str, int]]],
+    model: str | object | None,
+) -> Iterator[CustomModelStreamingResponse]:
+    """Convert the OpenAI ChatCompletionChunk response to CustomModelStreamingResponse."""
+    completion_id = str(uuid.uuid4())
+    created = int(time.time())
+
+    last_pipeline_interactions = None
+    last_usage_metrics = None
+
+    while True:
+        try:
+            (
+                response_text,
+                pipeline_interactions,
+                usage_metrics,
+            ) = next(streaming_response_iterator)
+            last_pipeline_interactions = pipeline_interactions
+            last_usage_metrics = usage_metrics
+
+            if response_text:
+                choice = ChunkChoice(
+                    index=0,
+                    delta=ChoiceDelta(role="assistant", content=response_text),
+                    finish_reason=None,
+                )
+                yield CustomModelStreamingResponse(
+                    id=completion_id,
+                    object="chat.completion.chunk",
+                    created=created,
+                    model=model,
+                    choices=[choice],
+                    usage=CompletionUsage(**usage_metrics) if usage_metrics else None,
+                )
+        except StopIteration:
+            break
+    # Yield final chunk indicating end of stream
+    choice = ChunkChoice(
+        index=0,
+        delta=ChoiceDelta(role="assistant"),
+        finish_reason="stop",
+    )
+    yield CustomModelStreamingResponse(
+        id=completion_id,
+        object="chat.completion.chunk",
+        created=created,
+        model=model,
+        choices=[choice],
+        usage=CompletionUsage(**last_usage_metrics) if last_usage_metrics else None,
+        pipeline_interactions=last_pipeline_interactions.model_dump_json()
+        if last_pipeline_interactions
+        else None,
+    )
+
+
+T = TypeVar("T")
+
+
+def async_gen_to_sync_thread(
+    async_iterator: AsyncIterator[T],
+    thread_pool_executor: ThreadPoolExecutor,
+    event_loop: asyncio.AbstractEventLoop,
+) -> Iterator[T]:
+    """Run an async iterator in a separate thread and provide a sync iterator."""
+    # A thread-safe queue for communication
+    sync_queue: queue.Queue[Any] = queue.Queue()
+    # A sentinel object to signal the end of the async generator
+    SENTINEL = object()  # noqa: N806
+
+    async def run_async_to_queue() -> None:
+        """Run in the separate thread's event loop."""
+        try:
+            async for item in async_iterator:
+                sync_queue.put(item)
+        except Exception as e:
+            # Put the exception on the queue to be re-raised in the main thread
+            sync_queue.put(e)
+        finally:
+            # Signal the end of iteration
+            sync_queue.put(SENTINEL)
+
+    thread_pool_executor.submit(event_loop.run_until_complete, run_async_to_queue()).result()
+
+    # The main thread consumes items synchronously
+    while True:
+        item = sync_queue.get()
+        if item is SENTINEL:
+            break
+        if isinstance(item, Exception):
+            raise item
+        yield item

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/src/datarobot_genai/core/custom_model.py

@@ -26,7 +26,6 @@ from concurrent.futures import ThreadPoolExecutor
 from typing import Any
 from typing import Literal
 
-from datarobot_drum import RuntimeParameters
 from openai.types.chat import CompletionCreateParams
 from openai.types.chat.completion_create_params import CompletionCreateParamsNonStreaming
 from openai.types.chat.completion_create_params import CompletionCreateParamsStreaming
@@ -41,26 +40,6 @@ from datarobot_genai.core.telemetry_agent import instrument
 logger = logging.getLogger(__name__)
 
 
-def maybe_set_env_from_runtime_parameters(key: str) -> None:
-    """Set an environment variable from a DRUM Runtime Parameter if it exists.
-
-    This is safe to call outside of the DataRobot runtime. If the parameter is not available,
-    the function does nothing.
-    """
-    runtime_parameter_placeholder_value = "SET_VIA_PULUMI_OR_MANUALLY"
-    try:
-        runtime_parameter_value = RuntimeParameters.get(key)
-        if (
-            runtime_parameter_value
-            and len(runtime_parameter_value) > 0
-            and runtime_parameter_value != runtime_parameter_placeholder_value
-        ):
-            os.environ[key] = runtime_parameter_value
-    except ValueError:
-        # Local dev: runtime parameters may be unavailable
-        pass
-
-
 def load_model() -> tuple[ThreadPoolExecutor, asyncio.AbstractEventLoop]:
     """Initialize a dedicated event loop within a worker thread.
 
@@ -83,7 +62,6 @@ def chat_entrypoint(
     load_model_result: tuple[ThreadPoolExecutor, asyncio.AbstractEventLoop],
     *,
     work_dir: str | None = None,
-    runtime_parameter_keys: list[str] | None = None,
     framework: Literal["crewai", "langgraph", "llamaindex", "nat"] | None = None,
     **kwargs: Any,
 ) -> CustomModelChatResponse | Iterator[CustomModelStreamingResponse]:
@@ -103,10 +81,6 @@ def chat_entrypoint(
     work_dir : Optional[str]
         Working directory to ``chdir`` into before invoking the agent. This is useful
         when relative paths are used in agent templates.
-    runtime_parameter_keys : Optional[List[str]]
-        Runtime parameter keys (DataRobot custom model) to propagate into env. When
-        ``None``, defaults to
-        ``['EXTERNAL_MCP_URL', 'MCP_DEPLOYMENT_ID']``.
     framework : Optional[Literal["crewai", "langgraph", "llamaindex", "nat"]]
         When provided, idempotently instruments HTTP clients, OpenAI SDK, and the
         given framework. If omitted, general instrumentation is still applied.
@@ -129,12 +103,6 @@ def chat_entrypoint(
         except Exception as e:
             logger.warning(f"Failed to change working directory to {work_dir}: {e}")
 
-    # Load MCP runtime parameters and session secret if configured
-    if runtime_parameter_keys is None:
-        runtime_parameter_keys = ["EXTERNAL_MCP_URL", "MCP_DEPLOYMENT_ID"]
-    for key in runtime_parameter_keys:
-        maybe_set_env_from_runtime_parameters(key)
-
     # Retrieve authorization context using all supported methods for downstream agents/tools
     completion_create_params["authorization_context"] = resolve_authorization_context(
         completion_create_params, **kwargs

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/src/datarobot_genai/core/utils/auth.py

@@ -74,7 +74,22 @@ class AuthContextHeaderHandler:
         if algorithm is None:
             raise ValueError("Algorithm None is not allowed. Use a secure algorithm like HS256.")
 
-        self.secret_key = secret_key or AuthContextConfig().session_secret_key
+        # Get secret key from parameter, config, or environment variable
+        # Handle the case where AuthContextConfig() initialization fails due to
+        # a bug in the datarobot package when SESSION_SECRET_KEY is not set
+        if secret_key:
+            self.secret_key = secret_key
+        else:
+            try:
+                config = AuthContextConfig()
+                self.secret_key = config.session_secret_key or ""
+            except (TypeError, AttributeError, Exception):
+                # Fallback to reading environment variable directly if config initialization fails
+                # This can happen when SESSION_SECRET_KEY is not set and the datarobot package's
+                # getenv function encounters a bug with None values
+                # it tries to check if "apiToken" in payload: when payload is None
+                self.secret_key = ""
+
         self.algorithm = algorithm
         self.validate_signature = validate_signature
 

{datarobot_genai-0.1.71 → datarobot_genai-0.2.0}/src/datarobot_genai/langgraph/agent.py

@@ -17,6 +17,15 @@ from collections.abc import AsyncGenerator
 from typing import Any
 from typing import cast
 
+from ag_ui.core import Event
+from ag_ui.core import EventType
+from ag_ui.core import TextMessageContentEvent
+from ag_ui.core import TextMessageEndEvent
+from ag_ui.core import TextMessageStartEvent
+from ag_ui.core import ToolCallArgsEvent
+from ag_ui.core import ToolCallEndEvent
+from ag_ui.core import ToolCallResultEvent
+from ag_ui.core import ToolCallStartEvent
 from langchain.tools import BaseTool
 from langchain_core.messages import AIMessageChunk
 from langchain_core.messages import ToolMessage
@@ -158,43 +167,7 @@ class LangGraphAgent(BaseAgent[BaseTool], abc.ABC):
         # The main difference is returning a generator for streaming or a final response for sync.
         if is_streaming(completion_create_params):
             # Streaming response: yield each message as it is generated
-            async def stream_generator() -> AsyncGenerator[
-                tuple[str, MultiTurnSample | None, UsageMetrics], None
-            ]:
-                # Iterate over the graph stream. For message events, yield the content.
-                # For update events, accumulate the usage metrics.
-                events = []
-                async for _, mode, event in graph_stream:
-                    if mode == "messages":
-                        message_event: tuple[AIMessageChunk, dict[str, Any]] = event  # type: ignore[assignment]
-                        llm_token, _ = message_event
-                        yield (
-                            str(llm_token.content),
-                            None,
-                            usage_metrics,
-                        )
-                    elif mode == "updates":
-                        update_event: dict[str, Any] = event  # type: ignore[assignment]
-                        events.append(update_event)
-                        current_node = next(iter(update_event))
-                        node_data = update_event[current_node]
-                        current_usage = node_data.get("usage", {}) if node_data is not None else {}
-                        if current_usage:
-                            usage_metrics["total_tokens"] += current_usage.get("total_tokens", 0)
-                            usage_metrics["prompt_tokens"] += current_usage.get("prompt_tokens", 0)
-                            usage_metrics["completion_tokens"] += current_usage.get(
-                                "completion_tokens", 0
-                            )
-                    else:
-                        raise ValueError(f"Invalid mode: {mode}")
-
-                # Create a list of events from the event listener
-                pipeline_interactions = self.create_pipeline_interactions_from_events(events)
-
-                # yield the final response indicating completion
-                yield "", pipeline_interactions, usage_metrics
-
-            return stream_generator()
+            return self._stream_generator(graph_stream, usage_metrics)
         else:
             # Synchronous response: collect all events and return the final message
             events: list[dict[str, Any]] = [
@@ -203,6 +176,16 @@ class LangGraphAgent(BaseAgent[BaseTool], abc.ABC):
                 if mode == "updates"
             ]
 
+            # Accumulate the usage metrics from the updates
+            for update in events:
+                current_node = next(iter(update))
+                node_data = update[current_node]
+                current_usage = node_data.get("usage", {}) if node_data is not None else {}
+                if current_usage:
+                    usage_metrics["total_tokens"] += current_usage.get("total_tokens", 0)
+                    usage_metrics["prompt_tokens"] += current_usage.get("prompt_tokens", 0)
+                    usage_metrics["completion_tokens"] += current_usage.get("completion_tokens", 0)
+
             pipeline_interactions = self.create_pipeline_interactions_from_events(events)
 
             # Extract the final event from the graph stream as the synchronous response
@@ -214,14 +197,132 @@ class LangGraphAgent(BaseAgent[BaseTool], abc.ABC):
                 if node_data is not None and "messages" in node_data
                 else ""
             )
-            current_usage = node_data.get("usage", {}) if node_data is not None else {}
-            if current_usage:
-                usage_metrics["total_tokens"] += current_usage.get("total_tokens", 0)
-                usage_metrics["prompt_tokens"] += current_usage.get("prompt_tokens", 0)
-                usage_metrics["completion_tokens"] += current_usage.get("completion_tokens", 0)
 
             return response_text, pipeline_interactions, usage_metrics
 
+    async def _stream_generator(
+        self, graph_stream: AsyncGenerator[tuple[Any, str, Any], None], usage_metrics: UsageMetrics
+    ) -> AsyncGenerator[tuple[str | Event, MultiTurnSample | None, UsageMetrics], None]:
+        # Iterate over the graph stream. For message events, yield the content.
+        # For update events, accumulate the usage metrics.
+        events = []
+        current_message_id = None
+        tool_call_id = ""
+        async for _, mode, event in graph_stream:
+            if mode == "messages":
+                message_event: tuple[AIMessageChunk | ToolMessage, dict[str, Any]] = event  # type: ignore[assignment]
+                message = message_event[0]
+                if isinstance(message, ToolMessage):
+                    yield (
+                        ToolCallEndEvent(
+                            type=EventType.TOOL_CALL_END, tool_call_id=message.tool_call_id
+                        ),
+                        None,
+                        usage_metrics,
+                    )
+                    yield (
+                        ToolCallResultEvent(
+                            type=EventType.TOOL_CALL_RESULT,
+                            message_id=message.id,
+                            tool_call_id=message.tool_call_id,
+                            content=message.content,
+                            role="tool",
+                        ),
+                        None,
+                        usage_metrics,
+                    )
+                    tool_call_id = ""
+                elif isinstance(message, AIMessageChunk):
+                    if message.tool_call_chunks:
+                        # This is a tool call message
+                        for tool_call_chunk in message.tool_call_chunks:
+                            if name := tool_call_chunk.get("name"):
+                                # Its a tool call start message
+                                tool_call_id = tool_call_chunk["id"]
+                                yield (
+                                    ToolCallStartEvent(
+                                        type=EventType.TOOL_CALL_START,
+                                        tool_call_id=tool_call_id,
+                                        tool_call_name=name,
+                                        parent_message_id=message.id,
+                                    ),
+                                    None,
+                                    usage_metrics,
+                                )
+                            elif args := tool_call_chunk.get("args"):
+                                # Its a tool call args message
+                                yield (
+                                    ToolCallArgsEvent(
+                                        type=EventType.TOOL_CALL_ARGS,
+                                        # Its empty when the tool chunk is not a start message
+                                        # So we use the tool call id from a previous start message
+                                        tool_call_id=tool_call_id,
+                                        delta=args,
+                                    ),
+                                    None,
+                                    usage_metrics,
+                                )
+                    elif message.content:
+                        # Its a text message
+                        # Handle the start and end of the text message
+                        if message.id != current_message_id:
+                            if current_message_id:
+                                yield (
+                                    TextMessageEndEvent(
+                                        type=EventType.TEXT_MESSAGE_END,
+                                        message_id=current_message_id,
+                                    ),
+                                    None,
+                                    usage_metrics,
+                                )
+                            current_message_id = message.id
+                            yield (
+                                TextMessageStartEvent(
+                                    type=EventType.TEXT_MESSAGE_START,
+                                    message_id=message.id,
+                                    role="assistant",
+                                ),
+                                None,
+                                usage_metrics,
+                            )
+                        yield (
+                            TextMessageContentEvent(
+                                type=EventType.TEXT_MESSAGE_CONTENT,
+                                message_id=message.id,
+                                delta=message.content,
+                            ),
+                            None,
+                            usage_metrics,
+                        )
+                else:
+                    raise ValueError(f"Invalid message event: {message_event}")
+            elif mode == "updates":
+                update_event: dict[str, Any] = event  # type: ignore[assignment]
+                events.append(update_event)
+                current_node = next(iter(update_event))
+                node_data = update_event[current_node]
+                current_usage = node_data.get("usage", {}) if node_data is not None else {}
+                if current_usage:
+                    usage_metrics["total_tokens"] += current_usage.get("total_tokens", 0)
+                    usage_metrics["prompt_tokens"] += current_usage.get("prompt_tokens", 0)
+                    usage_metrics["completion_tokens"] += current_usage.get("completion_tokens", 0)
+                if current_message_id:
+                    yield (
+                        TextMessageEndEvent(
+                            type=EventType.TEXT_MESSAGE_END,
+                            message_id=current_message_id,
+                        ),
+                        None,
+                        usage_metrics,
+                    )
+                    current_message_id = None
+
+        # Create a list of events from the event listener
+        pipeline_interactions = self.create_pipeline_interactions_from_events(events)
+
+        # yield the final response indicating completion
+        yield "", pipeline_interactions, usage_metrics
+
     @classmethod
     def create_pipeline_interactions_from_events(
         cls,