nvidia-nat-a2a 1.4.0a20251207__py3-none-any.whl

nat/plugins/a2a/client/client_impl.py

@@ -0,0 +1,296 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed 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.
+
+import logging
+from collections.abc import AsyncGenerator
+from typing import Any
+
+from pydantic import BaseModel
+from pydantic import Field
+
+from nat.builder.function import FunctionGroup
+from nat.builder.workflow_builder import Builder
+from nat.cli.register_workflow import register_function_group
+from nat.plugins.a2a.client.client_base import A2ABaseClient
+from nat.plugins.a2a.client.client_config import A2AClientConfig
+
+logger = logging.getLogger(__name__)
+
+
+# Input models for helper functions
+class GetTaskInput(BaseModel):
+    """Input for get_task function."""
+    task_id: str = Field(..., description="The ID of the task to retrieve")
+    history_length: int | None = Field(default=None, description="Number of history items to include")
+
+
+class CancelTaskInput(BaseModel):
+    """Input for cancel_task function."""
+    task_id: str = Field(..., description="The ID of the task to cancel")
+
+
+class SendMessageInput(BaseModel):
+    """Input for send_message function."""
+    query: str = Field(..., description="The query to send to the agent")
+    task_id: str | None = Field(default=None, description="Optional task ID for continuation")
+    context_id: str | None = Field(default=None, description="Optional context ID for session management")
+
+
+class A2AClientFunctionGroup(FunctionGroup):
+    """
+    A minimal FunctionGroup for A2A agents.
+
+    Exposes a simple `send_message` function to interact with A2A agents.
+    """
+
+    def __init__(self, config: A2AClientConfig, builder: Builder):
+        super().__init__(config=config)
+        self._builder = builder
+        self._client: A2ABaseClient | None = None
+        self._include_skills_in_description = config.include_skills_in_description
+
+    async def __aenter__(self):
+        """Initialize the A2A client and register functions."""
+        config: A2AClientConfig = self._config  # type: ignore[assignment]
+        base_url = str(config.url)
+
+        # Create and initialize A2A client
+        self._client = A2ABaseClient(base_url=base_url,
+                                     agent_card_path=config.agent_card_path,
+                                     task_timeout=config.task_timeout,
+                                     streaming=config.streaming)
+        await self._client.__aenter__()
+        logger.info("Connected to A2A agent at %s", base_url)
+
+        # Discover agent card and register functions
+        self._register_functions()
+
+        return self
+
+    def _register_functions(self):
+        """Retrieve agent card and register the three-level API: high-level, helpers, and low-level."""
+        # Validate client is initialized
+        if not self._client:
+            raise RuntimeError("A2A client not initialized")
+
+        # Get and validate agent card
+        agent_card = self._client.agent_card
+        if not agent_card:
+            raise RuntimeError("Agent card not available")
+
+        # Log agent information
+        logger.info("Agent: %s v%s", agent_card.name, agent_card.version)
+        if agent_card.skills:
+            logger.info("Skills: %s", [skill.name for skill in agent_card.skills])
+
+        # Register functions
+        # LEVEL 1: High-level main function (LLM-friendly)
+        self.add_function(
+            name="call",
+            fn=self._create_high_level_function(),
+            description=self._format_main_function_description(agent_card),
+        )
+
+        # LEVEL 2: Standard helpers (metadata/utility)
+        self.add_function(
+            name="get_skills",
+            fn=self._get_skills,
+            description="Get the list of skills and capabilities available from this agent",
+        )
+
+        self.add_function(
+            name="get_info",
+            fn=self._get_agent_info,
+            description="Get metadata about this agent (name, version, provider, capabilities)",
+        )
+
+        self.add_function(
+            name="get_task",
+            fn=self._wrap_get_task,
+            description="Get the status and details of a specific task by task_id",
+        )
+
+        self.add_function(
+            name="cancel_task",
+            fn=self._wrap_cancel_task,
+            description="Cancel a running task by task_id",
+        )
+
+        # LEVEL 3: Low-level protocol (advanced)
+        self.add_function(
+            name="send_message",
+            fn=self._send_message_advanced,
+            description=("Advanced: Send a message with full control over the A2A protocol. "
+                         "Returns raw events as a list. For most use cases, prefer using the "
+                         "high-level 'call()' function instead."),
+        )
+
+        self.add_function(
+            name="send_message_streaming",
+            fn=self._send_message_streaming,
+            description=("Advanced: Send a message and stream response events as they arrive. "
+                         "Yields raw events one by one. This is an async generator function. "
+                         "For most use cases, prefer using the high-level 'call()' function instead."),
+        )
+
+    async def __aexit__(self, exc_type, exc_value, traceback):
+        """Clean up the A2A client."""
+        if self._client:
+            await self._client.__aexit__(exc_type, exc_value, traceback)
+            self._client = None
+            logger.info("Disconnected from A2A agent")
+
+    def _format_main_function_description(self, agent_card) -> str:
+        """Create description for the main agent function."""
+        description = f"{agent_card.description}\n\n"
+
+        # Conditionally include skills based on configuration
+        if self._include_skills_in_description and agent_card.skills:
+            description += "**Capabilities:**\n"
+            for skill in agent_card.skills:
+                description += f"\n• **{skill.name}**: {skill.description}"
+                if skill.examples:
+                    examples = skill.examples[:2]  # Limit to 2 examples
+                    description += f"\n  Examples: {', '.join(examples)}"
+            description += "\n\n"
+        elif agent_card.skills:
+            # Brief mention that skills are available
+            description += f"**{len(agent_card.skills)} capabilities available.** "
+            description += "Use get_skills() for detailed information.\n\n"
+
+        description += "**Usage:** Send natural language queries to interact with this agent."
+
+        return description
+
+    def _create_high_level_function(self):
+        """High-level function that simplifies the response."""
+
+        async def high_level_fn(query: str, task_id: str | None = None, context_id: str | None = None) -> str:
+            """
+            Send a query to the agent and get a simple text response.
+
+            This is the recommended method for LLM usage.
+            For advanced use cases, use send_message() for raw events.
+            """
+            if not self._client:
+                raise RuntimeError("A2A client not initialized")
+
+            events = []
+            async for event in self._client.send_message(query, task_id, context_id):
+                events.append(event)
+
+            # Extract and return just the text response using base client helper
+            return self._client.extract_text_from_events(events)
+
+        return high_level_fn
+
+    async def _get_skills(self, params: dict | None = None) -> dict:
+        """Helper function to list agent skills."""
+        if not self._client or not self._client.agent_card:
+            return {"skills": []}
+
+        agent_card = self._client.agent_card
+        return {
+            "agent":
+                agent_card.name,
+            "skills": [{
+                "id": skill.id,
+                "name": skill.name,
+                "description": skill.description,
+                "examples": skill.examples or [],
+                "tags": skill.tags or []
+            } for skill in agent_card.skills]
+        }
+
+    async def _get_agent_info(self, params: dict | None = None) -> dict:
+        """Helper function to get agent metadata."""
+        if not self._client or not self._client.agent_card:
+            return {}
+
+        agent_card = self._client.agent_card
+        return {
+            "name": agent_card.name,
+            "description": agent_card.description,
+            "version": agent_card.version,
+            "provider": agent_card.provider.model_dump() if agent_card.provider else None,
+            "url": agent_card.url,
+            "capabilities": {
+                "streaming": agent_card.capabilities.streaming if agent_card.capabilities else False,
+            },
+            "num_skills": len(agent_card.skills)
+        }
+
+    async def _wrap_get_task(self, params: GetTaskInput) -> Any:
+        """Wrapper for get_task that delegates to client_base."""
+        if not self._client:
+            raise RuntimeError("A2A client not initialized")
+        return await self._client.get_task(params.task_id, params.history_length)
+
+    async def _wrap_cancel_task(self, params: CancelTaskInput) -> Any:
+        """Wrapper for cancel_task that delegates to client_base."""
+        if not self._client:
+            raise RuntimeError("A2A client not initialized")
+        return await self._client.cancel_task(params.task_id)
+
+    async def _send_message_advanced(self, params: SendMessageInput) -> list:
+        """
+        Send a message with full A2A protocol control.
+
+        Returns: List of ClientEvent|Message objects containing:
+        - Task information
+        - Status updates
+        - Artifact updates
+        - Full message details
+        """
+        if not self._client:
+            raise RuntimeError("A2A client not initialized")
+
+        events = []
+        async for event in self._client.send_message(params.query, params.task_id, params.context_id):
+            events.append(event)
+        return events
+
+    async def _send_message_streaming(self, params: SendMessageInput) -> AsyncGenerator[Any, None]:
+        """
+        Send a message with full A2A protocol control and stream events.
+
+        This is an async generator that yields events as they arrive from the agent.
+
+        Yields: ClientEvent|Message objects containing:
+        - Task information
+        - Status updates
+        - Artifact updates
+        - Full message details
+        """
+        if not self._client:
+            raise RuntimeError("A2A client not initialized")
+
+        async for event in self._client.send_message_streaming(params.query, params.task_id, params.context_id):
+            yield event
+
+
+@register_function_group(config_type=A2AClientConfig)
+async def a2a_client_function_group(config: A2AClientConfig, _builder: Builder):
+    """
+    Connect to an A2A agent, discover agent card and publish the primary
+    agent function and helper functions.
+
+    This function group creates a three-level API:
+    - High-level: Agent function named after the agent (e.g., dice_agent)
+    - Helpers: get_skills, get_info, get_task, cancel_task
+    - Low-level: send_message for advanced usage
+    """
+    async with A2AClientFunctionGroup(config, _builder) as group:
+        yield group

nat/plugins/a2a/register.py

@@ -0,0 +1,23 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed 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.
+
+# flake8: noqa
+# isort:skip_file
+
+# Register client components
+from .client import client_impl
+
+# Register server/frontend components
+from .server import register_frontend

nat/plugins/a2a/server/__init__.py

@@ -0,0 +1,14 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed 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.

nat/plugins/a2a/server/agent_executor_adapter.py

@@ -0,0 +1,172 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed 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.
+"""Adapter to bridge NAT workflows with A2A AgentExecutor interface.
+
+This module implements a message-only A2A agent for Phase 1, providing stateless
+request/response interactions without task lifecycle management.
+"""
+
+import logging
+
+from a2a.server.agent_execution import AgentExecutor
+from a2a.server.agent_execution import RequestContext
+from a2a.server.events import EventQueue
+from a2a.types import InternalError
+from a2a.types import InvalidParamsError
+from a2a.types import UnsupportedOperationError
+from a2a.utils import new_agent_text_message
+from a2a.utils.errors import ServerError
+from nat.runtime.session import SessionManager
+
+logger = logging.getLogger(__name__)
+
+
+class NATWorkflowAgentExecutor(AgentExecutor):
+    """Adapts NAT workflows to A2A AgentExecutor interface as a message-only agent.
+
+    This adapter implements Phase 1 support for A2A integration, providing stateless
+    message-based interactions. Each request is handled independently without maintaining
+    conversation state or task lifecycle.
+
+    Key characteristics:
+    - Stateless: Each message is processed independently
+    - Synchronous: Returns immediate responses (no long-running tasks)
+    - Message-only: Returns Message objects, not Task objects
+    - Concurrent: Uses SessionManager's semaphore for concurrency control
+
+    Note: Multi-turn conversations and task-based interactions are deferred to Phase 5.
+    """
+
+    def __init__(self, session_manager: SessionManager):
+        """Initialize the adapter with a NAT SessionManager.
+
+        Args:
+            session_manager: The SessionManager for handling workflow execution
+                with concurrency control via semaphore
+        """
+        self.session_manager = session_manager
+        logger.info("Initialized NATWorkflowAgentExecutor (message-only) for workflow: %s",
+                    session_manager.workflow.config.workflow.type)
+
+    async def execute(
+        self,
+        context: RequestContext,
+        event_queue: EventQueue,
+    ) -> None:
+        """Execute the NAT workflow and return a message response.
+
+        This is a message-only implementation (Phase 1):
+        1. Extracts the user query from the A2A message
+        2. Runs the NAT workflow (stateless, no conversation history)
+        3. Returns the result as a Message object (not a Task)
+
+        For Phase 1, each message is handled independently with no state preservation
+        between requests. The context_id and task_id from the A2A protocol are mapped
+        to NAT's conversation_id and user_message_id for tracing purposes only.
+
+        Args:
+            context: The A2A request context containing the user message
+            event_queue: Queue for sending the response message back to the client
+
+        Raises:
+            ServerError: If validation fails or workflow execution errors occur
+        """
+        # Validate the request
+        error = self._validate_request(context)
+        if error:
+            raise ServerError(error=InvalidParamsError())
+
+        # Extract query from the message
+        query = context.get_user_input()
+        if not query:
+            logger.error("No user input found in context")
+            raise ServerError(error=InvalidParamsError())
+
+        # Extract IDs for tracing (stored but not used for state management in Phase 1)
+        context_id = context.context_id
+        task_id = context.task_id
+
+        logger.info("Processing message-only request (context_id=%s, task_id=%s): %s", context_id, task_id, query[:100])
+
+        try:
+            # Run the NAT workflow using SessionManager for proper concurrency handling
+            # Each message gets its own independent session (stateless)
+            # TODO: Add support for user input callbacks and authentication in later phases
+            async with self.session_manager.session() as session:
+                async with session.run(query) as runner:
+                    # Get the result as a string
+                    response_text = await runner.result(to_type=str)
+
+            logger.info("Workflow completed successfully (context_id=%s, task_id=%s)", context_id, task_id)
+
+            # Create and send the response message (message-only pattern)
+            response_message = new_agent_text_message(
+                response_text,
+                context_id=context_id,
+                task_id=task_id,
+            )
+            await event_queue.enqueue_event(response_message)
+
+        except Exception as e:
+            logger.error("Error executing NAT workflow (context_id=%s, task_id=%s): %s",
+                         context_id,
+                         task_id,
+                         e,
+                         exc_info=True)
+
+            # Send error message back to client
+            error_message = new_agent_text_message(
+                f"An error occurred while processing your request: {str(e)}",
+                context_id=context_id,
+                task_id=task_id,
+            )
+            await event_queue.enqueue_event(error_message)
+            raise ServerError(error=InternalError()) from e
+
+    def _validate_request(self, context: RequestContext) -> bool:
+        """Validate the incoming request context.
+
+        Args:
+            context: The request context to validate
+
+        Returns:
+            True if validation fails, False if validation succeeds
+        """
+        # Basic validation - can be extended as needed
+        if not context.message:
+            logger.error("Request context has no message")
+            return True
+
+        return False
+
+    async def cancel(
+        self,
+        _context: RequestContext,
+        _event_queue: EventQueue,
+    ) -> None:
+        """Handle task cancellation requests.
+
+        Not applicable for message-only agents in Phase 1. Cancellation is a task-based
+        feature that will be implemented in Phase 5 along with long-running task support.
+
+        Args:
+            _context: The request context (unused in Phase 1)
+            _event_queue: Event queue for sending updates (unused in Phase 1)
+
+        Raises:
+            ServerError: Always raises UnsupportedOperationError
+        """
+        logger.warning("Task cancellation requested but not supported in message-only mode (Phase 1)")
+        raise ServerError(error=UnsupportedOperationError())

nat/plugins/a2a/server/front_end_config.py

@@ -0,0 +1,104 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed 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.
+
+import logging
+
+from pydantic import BaseModel
+from pydantic import Field
+
+from nat.data_models.front_end import FrontEndBaseConfig
+
+logger = logging.getLogger(__name__)
+
+
+class A2ACapabilitiesConfig(BaseModel):
+    """A2A agent capabilities configuration."""
+
+    streaming: bool = Field(
+        default=True,
+        description="Enable streaming responses (default: True)",
+    )
+    push_notifications: bool = Field(
+        default=False,
+        description="Enable push notifications (default: False)",
+    )
+
+
+class A2AFrontEndConfig(FrontEndBaseConfig, name="a2a"):
+    """A2A front end configuration.
+
+    A front end that exposes NeMo Agent toolkit workflows as A2A-compliant remote agents.
+    """
+
+    # Server settings
+    host: str = Field(
+        default="localhost",
+        description="Host to bind the server to (default: localhost)",
+    )
+    port: int = Field(
+        default=10000,
+        description="Port to bind the server to (default: 10000)",
+        ge=0,
+        le=65535,
+    )
+    version: str = Field(
+        default="1.0.0",
+        description="Version of the agent (default: 1.0.0)",
+    )
+    log_level: str = Field(
+        default="INFO",
+        description="Log level for the A2A server (default: INFO)",
+    )
+
+    # Agent metadata
+    name: str = Field(
+        default="NeMo Agent Toolkit A2A Agent",
+        description="Name of the A2A agent (default: NeMo Agent Toolkit A2A Agent)",
+    )
+    description: str = Field(
+        default="An AI agent powered by NeMo Agent Toolkit exposed via A2A protocol",
+        description="Description of what the agent does (default: generic description)",
+    )
+
+    # A2A capabilities
+    capabilities: A2ACapabilitiesConfig = Field(
+        default_factory=A2ACapabilitiesConfig,
+        description="Agent capabilities configuration",
+    )
+
+    # Concurrency control
+    max_concurrency: int = Field(
+        default=8,
+        description="Maximum number of concurrent workflow executions (default: 8). "
+        "Controls how many A2A requests can execute workflows simultaneously. "
+        "Set to 0 or -1 for unlimited concurrency.",
+        ge=-1,
+    )
+
+    # Content modes
+    default_input_modes: list[str] = Field(
+        default_factory=lambda: ["text", "text/plain"],
+        description="Supported input content types (default: text, text/plain)",
+    )
+    default_output_modes: list[str] = Field(
+        default_factory=lambda: ["text", "text/plain"],
+        description="Supported output content types (default: text, text/plain)",
+    )
+
+    # Optional customization
+    runner_class: str | None = Field(
+        default=None,
+        description="Custom worker class for handling A2A routes (default: built-in worker)",
+    )