Airflow Summit 2026 is coming August 31 - September 2 in Austin, TX. Register now to secure your spot!

Source code for airflow.providers.microsoft.azure.operators.ai_agents

#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
from __future__ import annotations

import time
from collections.abc import Sequence
from functools import cached_property
from typing import TYPE_CHECKING, Any, cast

from airflow.providers.common.compat.sdk import BaseOperator, conf
from airflow.providers.microsoft.azure._ai_agents import (
    VERSION_FAILURE_STATUSES,
    VERSION_INTERMEDIATE_STATUSES,
    VERSION_SUCCESS_STATUSES,
    _get_agent_version,
    _get_resource_attr,
    _get_version_status,
    _serialize_resource,
)
from airflow.providers.microsoft.azure.hooks.ai_agents import AzureAIAgentsHook
from airflow.providers.microsoft.azure.triggers.ai_agents import AzureAIAgentVersionTrigger

if TYPE_CHECKING:
    from azure.ai.projects.models import AgentBlueprintReference
    from pydantic import JsonValue

    from airflow.sdk import Context


[docs] class CreateAzureAIAgentOperator(BaseOperator): """ Create an Azure AI Hosted agent version from a container image definition. Creating a version also creates the Hosted agent itself on first use. :param agent_name: Hosted agent name. :param definition: Hosted agent definition. Include ``container_configuration.image`` with the Azure Container Registry image URI. :param metadata: Optional metadata attached to the Hosted agent. Default is ``None``. :param description: Optional human-readable Hosted agent description. Default is ``None``. :param blueprint_reference: Optional managed identity blueprint reference. Default is ``None``. :param wait_for_completion: Whether to wait until the created version reaches ``active``. Default is ``True``. :param poll_interval: Time in seconds between status checks. Default is ``30.0``. :param timeout: Time in seconds to wait for the version to become active. Default is ``3600``. :param deferrable: Run in deferrable mode when ``wait_for_completion`` is ``True``. Default is configured by ``operators.default_deferrable``. :param azure_ai_agents_conn_id: Azure AI Agents connection id. Default is ``azure_ai_agents_default``. :param endpoint: Optional Azure AI Foundry project endpoint override. Default is ``None``. :param api_version: Foundry Agent Service API version. Default is ``v1``. """
[docs] template_fields: Sequence[str] = ( "agent_name", "definition", "metadata", "description", "blueprint_reference", "azure_ai_agents_conn_id", "endpoint", "api_version", )
[docs] template_fields_renderers = { "definition": "json", "metadata": "json", "blueprint_reference": "json", }
[docs] ui_color = "#0678d4"
def __init__( self, *, agent_name: str, definition: dict[str, Any], metadata: dict[str, str] | None = None, description: str | None = None, blueprint_reference: AgentBlueprintReference | None = None, wait_for_completion: bool = True, poll_interval: float = 30.0, timeout: float = 60 * 60, deferrable: bool = conf.getboolean("operators", "default_deferrable", fallback=False), azure_ai_agents_conn_id: str = AzureAIAgentsHook.default_conn_name, endpoint: str | None = None, api_version: str = "v1", **kwargs, ) -> None: super().__init__(**kwargs)
[docs] self.agent_name = agent_name
[docs] self.definition = definition
[docs] self.metadata = metadata
[docs] self.description = description
[docs] self.blueprint_reference = blueprint_reference
[docs] self.wait_for_completion = wait_for_completion
[docs] self.poll_interval = poll_interval
[docs] self.timeout = timeout
[docs] self.deferrable = deferrable
[docs] self.azure_ai_agents_conn_id = azure_ai_agents_conn_id
[docs] self.endpoint = endpoint
[docs] self.api_version = api_version
@cached_property
[docs] def hook(self) -> AzureAIAgentsHook: """Create and return an AzureAIAgentsHook.""" return AzureAIAgentsHook( azure_ai_agents_conn_id=self.azure_ai_agents_conn_id, endpoint=self.endpoint, api_version=self.api_version, )
[docs] def execute(self, context: Context) -> dict[str, JsonValue]: """Create an Azure AI Hosted agent version and optionally wait for it to become active.""" self.log.info("Creating Azure AI Hosted agent %s version.", self.agent_name) version = self.hook.create_agent_version( agent_name=self.agent_name, definition=self.definition, metadata=self.metadata, description=self.description, blueprint_reference=self.blueprint_reference, ) if not self.wait_for_completion: return cast("dict[str, JsonValue]", _serialize_resource(version)) agent_version = _get_agent_version(version) if self.deferrable: self.defer( timeout=self.execution_timeout, trigger=AzureAIAgentVersionTrigger( azure_ai_agents_conn_id=self.azure_ai_agents_conn_id, endpoint=self.endpoint, api_version=self.api_version, agent_name=self.agent_name, agent_version=agent_version, timeout=self.timeout, poll_interval=self.poll_interval, ), method_name="execute_complete", ) return self._wait_for_version(agent_version=agent_version)
def _wait_for_version(self, *, agent_version: str) -> dict[str, JsonValue]: end_time = time.monotonic() + self.timeout while True: version = self.hook.get_agent_version(agent_name=self.agent_name, agent_version=agent_version) status = _get_version_status(version) if status in VERSION_SUCCESS_STATUSES: return cast("dict[str, JsonValue]", _serialize_resource(version)) if status in VERSION_FAILURE_STATUSES: error = _get_resource_attr(version, "error") or "No error details were returned" raise RuntimeError( f"Azure AI Hosted agent {self.agent_name} version {agent_version} failed: {error}." ) if status not in VERSION_INTERMEDIATE_STATUSES: raise RuntimeError( f"Azure AI Hosted agent {self.agent_name} version {agent_version} " f"reached unknown status {status}." ) if time.monotonic() >= end_time: raise TimeoutError( f"Timeout waiting for Azure AI Hosted agent {self.agent_name} version {agent_version}." ) self.log.info( "Azure AI Hosted agent %s version %s is in status %s. Sleeping for %s seconds.", self.agent_name, agent_version, status, self.poll_interval, ) time.sleep(self.poll_interval)
[docs] def execute_complete(self, context: Context, event: dict[str, Any] | None) -> dict[str, JsonValue]: """Resume after the version trigger completes.""" if event is None: raise RuntimeError("Trigger returned no event.") status = event.get("status") message = event.get("message", "No message returned from trigger.") if status == "success": if "version" not in event: raise RuntimeError("Trigger success event did not include version payload.") self.log.info(message) return event["version"] if status == "timeout": raise TimeoutError(message) if status == "error": raise RuntimeError(message) raise ValueError(f"Unexpected trigger event status: {status!r}")
[docs] class UpdateAzureAIAgentOperator(CreateAzureAIAgentOperator): """ Create a new version of an existing Azure AI Hosted agent. The Hosted agent API publishes updates as new immutable versions, so this operator accepts the same parameters as :class:`CreateAzureAIAgentOperator`. """
[docs] class RunAzureAIAgentOperator(BaseOperator): """ Invoke an Azure AI Hosted agent. :param agent_name: Hosted agent name. :param input_data: Request payload for the selected protocol. :param protocol: Hosted agent protocol to use. Supported values are ``responses`` and ``invocations``. Default is ``responses``. :param agent_session_id: Optional session id used by the ``invocations`` protocol. Default is ``None``. :param user_isolation_key: Optional per-user isolation key for endpoint-scoped resources. Default is ``None``. :param azure_ai_agents_conn_id: Azure AI Agents connection id. Default is ``azure_ai_agents_default``. :param endpoint: Optional Azure AI Foundry project endpoint override. Default is ``None``. :param api_version: Foundry Agent Service API version. Default is ``v1``. """
[docs] template_fields: Sequence[str] = ( "agent_name", "input_data", "protocol", "agent_session_id", "user_isolation_key", "azure_ai_agents_conn_id", "endpoint", "api_version", )
[docs] template_fields_renderers = {"input_data": "json"}
[docs] ui_color = "#0678d4"
def __init__( self, *, agent_name: str, input_data: dict[str, Any], protocol: str = "responses", agent_session_id: str | None = None, user_isolation_key: str | None = None, azure_ai_agents_conn_id: str = AzureAIAgentsHook.default_conn_name, endpoint: str | None = None, api_version: str = "v1", **kwargs, ) -> None: super().__init__(**kwargs)
[docs] self.agent_name = agent_name
[docs] self.input_data = input_data
[docs] self.protocol = protocol
[docs] self.agent_session_id = agent_session_id
[docs] self.user_isolation_key = user_isolation_key
[docs] self.azure_ai_agents_conn_id = azure_ai_agents_conn_id
[docs] self.endpoint = endpoint
[docs] self.api_version = api_version
@cached_property
[docs] def hook(self) -> AzureAIAgentsHook: """Create and return an AzureAIAgentsHook.""" return AzureAIAgentsHook( azure_ai_agents_conn_id=self.azure_ai_agents_conn_id, endpoint=self.endpoint, api_version=self.api_version, )
[docs] def execute(self, context: Context) -> JsonValue: """Invoke an Azure AI Hosted agent and return the response payload.""" self.log.info("Invoking Azure AI Hosted agent %s with %s protocol.", self.agent_name, self.protocol) if self.protocol == "responses": return self.hook.invoke_agent_responses( agent_name=self.agent_name, input_data=self.input_data, user_isolation_key=self.user_isolation_key, ) if self.protocol == "invocations": return self.hook.invoke_agent_invocations( agent_name=self.agent_name, input_data=self.input_data, agent_session_id=self.agent_session_id, user_isolation_key=self.user_isolation_key, ) raise ValueError("protocol must be either 'responses' or 'invocations'.")
[docs] class DeleteAzureAIAgentOperator(BaseOperator): """ Delete an Azure AI Hosted agent or one Hosted agent version. :param agent_name: Hosted agent name. :param agent_version: Optional Hosted agent version. When omitted, the whole agent is deleted. Default is ``None``. :param force: Whether to cascade-delete active sessions when deleting the agent or version. Default is ``False``. :param wait_for_completion: Whether to wait until the resource is deleted. Default is ``True``. :param poll_interval: Time in seconds between status checks. Default is ``30.0``. :param timeout: Time in seconds to wait for deletion to complete. Default is ``3600``. :param azure_ai_agents_conn_id: Azure AI Agents connection id. Default is ``azure_ai_agents_default``. :param endpoint: Optional Azure AI Foundry project endpoint override. Default is ``None``. :param api_version: Foundry Agent Service API version. Default is ``v1``. """
[docs] template_fields: Sequence[str] = ( "agent_name", "agent_version", "azure_ai_agents_conn_id", "endpoint", "api_version", )
[docs] ui_color = "#0678d4"
def __init__( self, *, agent_name: str, agent_version: str | None = None, force: bool = False, wait_for_completion: bool = True, poll_interval: float = 30.0, timeout: float = 60 * 60, azure_ai_agents_conn_id: str = AzureAIAgentsHook.default_conn_name, endpoint: str | None = None, api_version: str = "v1", **kwargs, ) -> None: super().__init__(**kwargs)
[docs] self.agent_name = agent_name
[docs] self.agent_version = agent_version
[docs] self.force = force
[docs] self.wait_for_completion = wait_for_completion
[docs] self.poll_interval = poll_interval
[docs] self.timeout = timeout
[docs] self.azure_ai_agents_conn_id = azure_ai_agents_conn_id
[docs] self.endpoint = endpoint
[docs] self.api_version = api_version
@cached_property
[docs] def hook(self) -> AzureAIAgentsHook: """Create and return an AzureAIAgentsHook.""" return AzureAIAgentsHook( azure_ai_agents_conn_id=self.azure_ai_agents_conn_id, endpoint=self.endpoint, api_version=self.api_version, )
[docs] def execute(self, context: Context) -> dict[str, JsonValue]: """Delete an Azure AI Hosted agent or version and optionally wait for deletion.""" delete_response: Any if self.agent_version is None: self.log.info("Deleting Azure AI Hosted agent %s.", self.agent_name) delete_response = self.hook.delete_agent(agent_name=self.agent_name, force=self.force) else: self.log.info( "Deleting Azure AI Hosted agent %s version %s.", self.agent_name, self.agent_version ) delete_response = self.hook.delete_agent_version( agent_name=self.agent_name, agent_version=self.agent_version, force=self.force ) if self.wait_for_completion: self._wait_for_deletion() return cast("dict[str, JsonValue]", _serialize_resource(delete_response))
def _wait_for_deletion(self) -> None: end_time = time.monotonic() + self.timeout resource_description = f"agent {self.agent_name}" if self.agent_version is not None: resource_description += f" version {self.agent_version}" while True: if self.agent_version is None: is_deleted = self.hook.is_agent_deleted(agent_name=self.agent_name) else: is_deleted = self.hook.is_agent_version_deleted( agent_name=self.agent_name, agent_version=self.agent_version ) if is_deleted: self.log.info("Azure AI Hosted %s was deleted.", resource_description) return if time.monotonic() >= end_time: raise TimeoutError(f"Timeout waiting for Azure AI Hosted {resource_description} deletion.") self.log.info( "Azure AI Hosted %s is still retrievable. Sleeping for %s seconds.", resource_description, self.poll_interval, ) time.sleep(self.poll_interval)

Was this entry helpful?