yujunjun/autogen
1
0
Fork 0
mirror of https://github.com/microsoft/autogen.git synced 2025-08-21 23:22:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add tool_agent_caller_loop and group chat notebook. (#405)

Browse Source 
* Add tool_agent_caller_loop and group chat notebook.

* Fix types

* fix ref

---------

Co-authored-by: Jack Gerrits <jackgerrits@users.noreply.github.com>
This commit is contained in:
Eric Zhu 2024-08-27 12:11:48 -07:00 committed by GitHub
parent c8f6f3bb38
commit 12cf331e71
9 changed files with 924 additions and 354 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
python/docs/src/core-concepts/overview.md
View File

@ -116,7 +116,7 @@ implementation of the contracts determines how agents handle messages.
The behavior contract is sometimes referred to as the message protocol. The behavior contract is sometimes referred to as the message protocol.
It is the developer's responsibility to implement the behavior contract. It is the developer's responsibility to implement the behavior contract.
Multi-agent patterns are design patterns that emerge from behavior contracts Multi-agent patterns are design patterns that emerge from behavior contracts
(see [Multi-Agent Design Patterns](../getting-started/multi-agent-design-patterns.ipynb)). (see [Multi-Agent Design Patterns](../getting-started/multi-agent-design-patterns.md)).
### An Example Application ### An Example Application

399
python/docs/src/getting-started/group-chat.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

18
python/docs/src/getting-started/multi-agent-design-patterns.md Normal file
View File

@ -0,0 +1,18 @@
# Multi-Agent Design Patterns
Agents can work together in a variety of ways to solve problems.
Research works like [AutoGen](https://aka.ms/autogen-paper),
[MetaGPT](https://arxiv.org/abs/2308.00352)
and [ChatDev](https://arxiv.org/abs/2307.07924) have shown
multi-agent systems out-performing single agent systems at complex tasks
like software development.
A multi-agent design pattern is a structure that emerges from message protocols:
it describes how agents interact with each other to solve problems.
For example, the [tool-equiped agent](./tools.ipynb#tool-equipped-agent) in
the previous section employs a design pattern called ReAct,
which involves an agent interacting with tools.
You can implement any multi-agent design pattern using AGNext agents.
In the next two sections, we will discuss two common design patterns:
group chat for task decomposition, and reflection for robustness.

33
python/docs/src/getting-started/multi-agent-design-patterns.ipynb → python/docs/src/getting-started/reflection.ipynb
View File

@ -4,30 +4,7 @@
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
"source": [ "source": [
"# Multi-Agent Design Patterns\n", "# Reflection\n",
"\n",
"Agents can work together in a variety of ways to solve problems.\n",
"Research works like [AutoGen](https://aka.ms/autogen-paper),\n",
"[MetaGPT](https://arxiv.org/abs/2308.00352)\n",
"and [ChatDev](https://arxiv.org/abs/2307.07924) have shown\n",
"multi-agent systems out-performing single agent systems at complex tasks\n",
"like software development.\n",
"\n",
"A multi-agent design pattern is a structure that emerges from message protocols:\n",
"it describes how agents interact with each other to solve problems.\n",
"For example, the [tool-equiped agent](./tools.ipynb#tool-equipped-agent) in\n",
"the previous section employs a design pattern called ReAct,\n",
"which involves an agent interacting with tools.\n",
"\n",
"You can implement any multi-agent design pattern using AGNext agents.\n",
"In this section, we use the reflection pattern as an example."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Reflection\n",
"\n", "\n",
"Reflection is a design pattern where an LLM generation is followed by a reflection,\n", "Reflection is a design pattern where an LLM generation is followed by a reflection,\n",
"which in itself is another LLM generation conditioned on the output of the first one.\n", "which in itself is another LLM generation conditioned on the output of the first one.\n",
@ -50,7 +27,7 @@
"will generate a code snippet, and the reviewer agent will generate a critique\n", "will generate a code snippet, and the reviewer agent will generate a critique\n",
"of the code snippet.\n", "of the code snippet.\n",
"\n", "\n",
"### Message Protocol\n", "## Message Protocol\n",
"\n", "\n",
"Before we define the agents, we need to first define the message protocol for the agents." "Before we define the agents, we need to first define the message protocol for the agents."
] ]
@ -107,7 +84,7 @@
"\n", "\n",
"![coder-reviewer data flow](coder-reviewer-data-flow.svg)\n", "![coder-reviewer data flow](coder-reviewer-data-flow.svg)\n",
"\n", "\n",
"### Agents\n", "## Agents\n",
"\n", "\n",
"Now, let's define the agents for the reflection design pattern." "Now, let's define the agents for the reflection design pattern."
] ]
@ -376,7 +353,7 @@
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
"source": [ "source": [
"### Logging\n", "## Logging\n",
"\n", "\n",
"Turn on logging to see the messages exchanged between the agents." "Turn on logging to see the messages exchanged between the agents."
] ]
@ -397,7 +374,7 @@
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
"source": [ "source": [
"### Running the Design Pattern\n", "## Running the Design Pattern\n",
"\n", "\n",
"Let's test the design pattern with a coding task." "Let's test the design pattern with a coding task."
] ]

640
python/docs/src/getting-started/tools.ipynb
View File

@ -1,324 +1,320 @@
{ {
"cells": [ "cells": [
{ {
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
"source": [ "source": [
"# Tools\n", "# Tools\n",
"\n", "\n",
"Tools are code that can be executed by an agent to perform actions. A tool\n", "Tools are code that can be executed by an agent to perform actions. A tool\n",
"can be a simple function such as a calculator, or an API call to a third-party service\n", "can be a simple function such as a calculator, or an API call to a third-party service\n",
"such as stock price lookup and weather forecast.\n", "such as stock price lookup and weather forecast.\n",
"In the context of AI agents, tools are designed to be executed by agents in\n", "In the context of AI agents, tools are designed to be executed by agents in\n",
"response to model-generated function calls.\n", "response to model-generated function calls.\n",
"\n", "\n",
"AGNext provides the {py:mod}`agnext.components.tools` module with a suite of built-in\n", "AGNext provides the {py:mod}`agnext.components.tools` module with a suite of built-in\n",
"tools and utilities for creating and running custom tools." "tools and utilities for creating and running custom tools."
] ]
}, },
{ {
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
"source": [ "source": [
"## Built-in Tools\n", "## Built-in Tools\n",
"\n", "\n",
"One of the built-in tools is the {py:class}`agnext.components.tools.PythonCodeExecutionTool`,\n", "One of the built-in tools is the {py:class}`agnext.components.tools.PythonCodeExecutionTool`,\n",
"which allows agents to execute Python code snippets.\n", "which allows agents to execute Python code snippets.\n",
"\n", "\n",
"Here is how you create the tool and use it." "Here is how you create the tool and use it."
] ]
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": null, "execution_count": 1,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [
"source": [ {
"from agnext.components.code_executor import LocalCommandLineCodeExecutor\n", "name": "stdout",
"from agnext.components.tools import PythonCodeExecutionTool\n", "output_type": "stream",
"from agnext.core import CancellationToken\n", "text": [
"\n", "Hello, world!\n",
"# Create the tool.\n", "\n"
"code_executor = LocalCommandLineCodeExecutor()\n", ]
"code_execution_tool = PythonCodeExecutionTool(code_executor)\n", }
"cancellation_token = CancellationToken()\n", ],
"\n", "source": [
"# Use the tool directly without an agent.\n", "from agnext.components.code_executor import LocalCommandLineCodeExecutor\n",
"code = \"print('Hello, world!')\"\n", "from agnext.components.tools import PythonCodeExecutionTool\n",
"result = await code_execution_tool.run_json({\"code\": code}, cancellation_token)\n", "from agnext.core import CancellationToken\n",
"print(code_execution_tool.return_value_as_string(result))" "\n",
] "# Create the tool.\n",
}, "code_executor = LocalCommandLineCodeExecutor()\n",
{ "code_execution_tool = PythonCodeExecutionTool(code_executor)\n",
"cell_type": "markdown", "cancellation_token = CancellationToken()\n",
"metadata": {}, "\n",
"source": [ "# Use the tool directly without an agent.\n",
"The {py:class}`~agnext.components.code_executor.LocalCommandLineCodeExecutor`\n", "code = \"print('Hello, world!')\"\n",
"class is a built-in code executor that runs Python code snippets in a subprocess\n", "result = await code_execution_tool.run_json({\"code\": code}, cancellation_token)\n",
"in the local command line environment.\n", "print(code_execution_tool.return_value_as_string(result))"
"The {py:class}`~agnext.components.tools.PythonCodeExecutionTool` class wraps the code executor\n", ]
"and provides a simple interface to execute Python code snippets.\n", },
"\n", {
"Other built-in tools will be added in the future." "cell_type": "markdown",
] "metadata": {},
}, "source": [
{ "The {py:class}`~agnext.components.code_executor.LocalCommandLineCodeExecutor`\n",
"cell_type": "markdown", "class is a built-in code executor that runs Python code snippets in a subprocess\n",
"metadata": {}, "in the local command line environment.\n",
"source": [ "The {py:class}`~agnext.components.tools.PythonCodeExecutionTool` class wraps the code executor\n",
"## Custom Function Tools\n", "and provides a simple interface to execute Python code snippets.\n",
"\n", "\n",
"A tool can also be a simple Python function that performs a specific action.\n", "Other built-in tools will be added in the future."
"To create a custom function tool, you just need to create a Python function\n", ]
"and use the {py:class}`agnext.components.tools.FunctionTool` class to wrap it.\n", },
"\n", {
"For example, a simple tool to obtain the stock price of a company might look like this:" "cell_type": "markdown",
] "metadata": {},
}, "source": [
{ "## Custom Function Tools\n",
"cell_type": "code", "\n",
"execution_count": 5, "A tool can also be a simple Python function that performs a specific action.\n",
"metadata": {}, "To create a custom function tool, you just need to create a Python function\n",
"outputs": [ "and use the {py:class}`agnext.components.tools.FunctionTool` class to wrap it.\n",
{ "\n",
"name": "stdout", "For example, a simple tool to obtain the stock price of a company might look like this:"
"output_type": "stream", ]
"text": [ },
"138.75280591295171\n" {
] "cell_type": "code",
} "execution_count": 2,
], "metadata": {},
"source": [ "outputs": [
"import random\n", {
"\n", "name": "stdout",
"from agnext.components.tools import FunctionTool\n", "output_type": "stream",
"from agnext.core import CancellationToken\n", "text": [
"from typing_extensions import Annotated\n", "194.71306528148511\n"
"\n", ]
"\n", }
"async def get_stock_price(ticker: str, date: Annotated[str, \"Date in YYYY/MM/DD\"]) -> float:\n", ],
" # Returns a random stock price for demonstration purposes.\n", "source": [
" return random.uniform(10, 200)\n", "import random\n",
"\n", "\n",
"\n", "from agnext.components.tools import FunctionTool\n",
"# Create a function tool.\n", "from agnext.core import CancellationToken\n",
"stock_price_tool = FunctionTool(get_stock_price, description=\"Get the stock price.\")\n", "from typing_extensions import Annotated\n",
"\n", "\n",
"# Run the tool.\n", "\n",
"cancellation_token = CancellationToken()\n", "async def get_stock_price(ticker: str, date: Annotated[str, \"Date in YYYY/MM/DD\"]) -> float:\n",
"result = await stock_price_tool.run_json({\"ticker\": \"AAPL\", \"date\": \"2021/01/01\"}, cancellation_token)\n", " # Returns a random stock price for demonstration purposes.\n",
"\n", " return random.uniform(10, 200)\n",
"# Print the result.\n", "\n",
"print(stock_price_tool.return_value_as_string(result))" "\n",
] "# Create a function tool.\n",
}, "stock_price_tool = FunctionTool(get_stock_price, description=\"Get the stock price.\")\n",
{ "\n",
"cell_type": "markdown", "# Run the tool.\n",
"metadata": {}, "cancellation_token = CancellationToken()\n",
"source": [ "result = await stock_price_tool.run_json({\"ticker\": \"AAPL\", \"date\": \"2021/01/01\"}, cancellation_token)\n",
"## Tool-Equipped Agent\n", "\n",
"\n", "# Print the result.\n",
"To use tools with an agent, you can use {py:class}`agnext.components.tool_agent.ToolAgent`,\n", "print(stock_price_tool.return_value_as_string(result))"
"by using it in a composition pattern.\n", ]
"Here is an example tool-use agent that uses {py:class}`~agnext.components.tool_agent.ToolAgent`\n", },
"as an inner agent for executing tools." {
] "cell_type": "markdown",
}, "metadata": {},
{ "source": [
"cell_type": "code", "## Tool-Equipped Agent\n",
"execution_count": 6, "\n",
"metadata": {}, "To use tools with an agent, you can use {py:class}`agnext.components.tool_agent.ToolAgent`,\n",
"outputs": [], "by using it in a composition pattern.\n",
"source": [ "Here is an example tool-use agent that uses {py:class}`~agnext.components.tool_agent.ToolAgent`\n",
"import asyncio\n", "as an inner agent for executing tools."
"from dataclasses import dataclass\n", ]
"from typing import List\n", },
"\n", {
"from agnext.application import SingleThreadedAgentRuntime\n", "cell_type": "code",
"from agnext.components import FunctionCall, RoutedAgent, message_handler\n", "execution_count": 3,
"from agnext.components.models import (\n", "metadata": {},
" AssistantMessage,\n", "outputs": [],
" ChatCompletionClient,\n", "source": [
" FunctionExecutionResult,\n", "from dataclasses import dataclass\n",
" FunctionExecutionResultMessage,\n", "from typing import List\n",
" LLMMessage,\n", "\n",
" OpenAIChatCompletionClient,\n", "from agnext.application import SingleThreadedAgentRuntime\n",
" SystemMessage,\n", "from agnext.components import RoutedAgent, message_handler\n",
" UserMessage,\n", "from agnext.components.models import (\n",
")\n", " ChatCompletionClient,\n",
"from agnext.components.tool_agent import ToolAgent, ToolException\n", " LLMMessage,\n",
"from agnext.components.tools import FunctionTool, Tool, ToolSchema\n", " OpenAIChatCompletionClient,\n",
"from agnext.core import AgentId, AgentInstantiationContext, MessageContext\n", " SystemMessage,\n",
"\n", " UserMessage,\n",
"\n", ")\n",
"@dataclass\n", "from agnext.components.tool_agent import ToolAgent, tool_agent_caller_loop\n",
"class Message:\n", "from agnext.components.tools import FunctionTool, Tool, ToolSchema\n",
" content: str\n", "from agnext.core import AgentId, AgentInstantiationContext, MessageContext\n",
"\n", "\n",
"\n", "\n",
"class ToolUseAgent(RoutedAgent):\n", "@dataclass\n",
" def __init__(self, model_client: ChatCompletionClient, tool_schema: List[ToolSchema], tool_agent: AgentId) -> None:\n", "class Message:\n",
" super().__init__(\"An agent with tools\")\n", " content: str\n",
" self._system_messages: List[LLMMessage] = [SystemMessage(\"You are a helpful AI assistant.\")]\n", "\n",
" self._model_client = model_client\n", "\n",
" self._tool_schema = tool_schema\n", "class ToolUseAgent(RoutedAgent):\n",
" self._tool_agent = tool_agent\n", " def __init__(self, model_client: ChatCompletionClient, tool_schema: List[ToolSchema], tool_agent: AgentId) -> None:\n",
"\n", " super().__init__(\"An agent with tools\")\n",
" @message_handler\n", " self._system_messages: List[LLMMessage] = [SystemMessage(\"You are a helpful AI assistant.\")]\n",
" async def handle_user_message(self, message: Message, ctx: MessageContext) -> Message:\n", " self._model_client = model_client\n",
" # Create a session of messages.\n", " self._tool_schema = tool_schema\n",
" session: List[LLMMessage] = [UserMessage(content=message.content, source=\"user\")]\n", " self._tool_agent = tool_agent\n",
" # Get a response from the model.\n", "\n",
" response = await self._model_client.create(\n", " @message_handler\n",
" self._system_messages + session, tools=self._tool_schema, cancellation_token=cancellation_token\n", " async def handle_user_message(self, message: Message, ctx: MessageContext) -> Message:\n",
" )\n", " # Create a session of messages.\n",
" # Add the response to the session.\n", " session: List[LLMMessage] = [UserMessage(content=message.content, source=\"user\")]\n",
" session.append(AssistantMessage(content=response.content, source=\"assistant\"))\n", " # Run the caller loop to handle tool calls.\n",
"\n", " messages = await tool_agent_caller_loop(\n",
" # Keep iterating until the model stops generating tool calls.\n", " self,\n",
" while isinstance(response.content, list) and all(isinstance(item, FunctionCall) for item in response.content):\n", " tool_agent_id=self._tool_agent,\n",
" # Execute functions called by the model by sending messages to itself.\n", " model_client=self._model_client,\n",
" results: List[FunctionExecutionResult | BaseException] = await asyncio.gather(\n", " input_messages=session,\n",
" *[self.send_message(call, self._tool_agent) for call in response.content],\n", " tool_schema=self._tool_schema,\n",
" return_exceptions=True,\n", " cancellation_token=ctx.cancellation_token,\n",
" )\n", " )\n",
" # Combine the results into a single response and handle exceptions.\n", " # Return the final response.\n",
" function_results: List[FunctionExecutionResult] = []\n", " assert isinstance(messages[-1].content, str)\n",
" for result in results:\n", " return Message(content=messages[-1].content)"
" if isinstance(result, FunctionExecutionResult):\n", ]
" function_results.append(result)\n", },
" elif isinstance(result, ToolException):\n", {
" function_results.append(FunctionExecutionResult(content=f\"Error: {result}\", call_id=result.call_id))\n", "cell_type": "markdown",
" elif isinstance(result, BaseException):\n", "metadata": {},
" raise result # Unexpected exception.\n", "source": [
" session.append(FunctionExecutionResultMessage(content=function_results))\n", "The `ToolUseAgent` class uses a convenience function {py:meth}`agnext.components.tool_agent.tool_agent_caller_loop`, \n",
" # Query the model again with the new response.\n", "to handle the interaction between the model and the tool agent.\n",
" response = await self._model_client.create(\n", "The core idea can be described using a simple control flow graph:\n",
" self._system_messages + session, tools=self._tool_schema, cancellation_token=cancellation_token\n", "\n",
" )\n", "![ToolUseAgent control flow graph](tool-use-agent-cfg.svg)\n",
" session.append(AssistantMessage(content=response.content, source=self.metadata[\"type\"]))\n", "\n",
"\n", "The `ToolUseAgent`'s `handle_user_message` handler handles messages from the user,\n",
" # Return the final response.\n", "and determines whether the model has generated a tool call.\n",
" assert isinstance(response.content, str)\n", "If the model has generated tool calls, then the handler sends a function call\n",
" return Message(content=response.content)" "message to the {py:class}`~agnext.components.tool_agent.ToolAgent` agent\n",
] "to execute the tools,\n",
}, "and then queries the model again with the results of the tool calls.\n",
{ "This process continues until the model stops generating tool calls,\n",
"cell_type": "markdown", "at which point the final response is returned to the user.\n",
"metadata": {}, "\n",
"source": [ "By having the tool execution logic in a separate agent,\n",
"The `ToolUseAgent` class is a bit involved, however,\n", "we expose the model-tool interactions to the agent runtime as messages, so the tool executions\n",
"the core idea can be described using a simple control flow graph:\n", "can be observed externally and intercepted if necessary.\n",
"\n", "\n",
"![ToolUseAgent control flow graph](tool-use-agent-cfg.svg)\n", "To run the agent, we need to create a runtime and register the agent."
"\n", ]
"The `ToolUseAgent`'s `handle_user_message` handler handles messages from the user,\n", },
"and determines whether the model has generated a tool call.\n", {
"If the model has generated tool calls, then the handler sends a function call\n", "cell_type": "code",
"message to the {py:class}`~agnext.components.tool_agent.ToolAgent` agent\n", "execution_count": 4,
"to execute the tools,\n", "metadata": {},
"and then queries the model again with the results of the tool calls.\n", "outputs": [
"This process continues until the model stops generating tool calls,\n", {
"at which point the final response is returned to the user.\n", "data": {
"\n", "text/plain": [
"By having the tool execution logic in a separate agent,\n", "AgentType(type='tool_use_agent')"
"we expose the model-tool interactions to the agent runtime as messages, so the tool executions\n", ]
"can be observed externally and intercepted if necessary.\n", },
"\n", "execution_count": 4,
"To run the agent, we need to create a runtime and register the agent." "metadata": {},
] "output_type": "execute_result"
}, }
{ ],
"cell_type": "code", "source": [
"execution_count": 7, "# Create a runtime.\n",
"metadata": {}, "runtime = SingleThreadedAgentRuntime()\n",
"outputs": [], "# Create the tools.\n",
"source": [ "tools: List[Tool] = [FunctionTool(get_stock_price, description=\"Get the stock price.\")]\n",
"# Create a runtime.\n", "# Register the agents.\n",
"runtime = SingleThreadedAgentRuntime()\n", "await runtime.register(\n",
"# Create the tools.\n", " \"tool_executor_agent\",\n",
"tools: List[Tool] = [FunctionTool(get_stock_price, description=\"Get the stock price.\")]\n", " lambda: ToolAgent(\n",
"# Register the agents.\n", " description=\"Tool Executor Agent\",\n",
"await runtime.register(\n", " tools=tools,\n",
" \"tool-executor-agent\",\n", " ),\n",
" lambda: ToolAgent(\n", ")\n",
" description=\"Tool Executor Agent\",\n", "await runtime.register(\n",
" tools=tools,\n", " \"tool_use_agent\",\n",
" ),\n", " lambda: ToolUseAgent(\n",
")\n", " OpenAIChatCompletionClient(model=\"gpt-4o-mini\"),\n",
"await runtime.register(\n", " tool_schema=[tool.schema for tool in tools],\n",
" \"tool-use-agent\",\n", " tool_agent=AgentId(\"tool_executor_agent\", AgentInstantiationContext.current_agent_id().key),\n",
" lambda: ToolUseAgent(\n", " ),\n",
" OpenAIChatCompletionClient(model=\"gpt-4o-mini\"),\n", ")"
" tool_schema=[tool.schema for tool in tools],\n", ]
" tool_agent=AgentId(\"tool-executor-agent\", AgentInstantiationContext.current_agent_id().key),\n", },
" ),\n", {
")" "cell_type": "markdown",
] "metadata": {},
}, "source": [
{ "This example uses the {py:class}`agnext.components.models.OpenAIChatCompletionClient`,\n",
"cell_type": "markdown", "for Azure OpenAI and other clients, see [Model Clients](./model-clients.ipynb).\n",
"metadata": {}, "Let's test the agent with a question about stock price."
"source": [ ]
"This example uses the {py:class}`agnext.components.models.OpenAIChatCompletionClient`,\n", },
"for Azure OpenAI and other clients, see [Model Clients](./model-clients.ipynb).\n", {
"Let's test the agent with a question about stock price." "cell_type": "code",
] "execution_count": 5,
}, "metadata": {},
{ "outputs": [
"cell_type": "code", {
"execution_count": 8, "name": "stdout",
"metadata": {}, "output_type": "stream",
"outputs": [ "text": [
{ "The stock price of NVIDIA (NVDA) on June 1, 2024, was approximately $148.86.\n"
"name": "stdout", ]
"output_type": "stream", }
"text": [ ],
"The stock price of NVDA on June 1, 2024, is approximately $49.28.\n" "source": [
] "# Start processing messages.\n",
} "runtime.start()\n",
], "# Send a direct message to the tool agent.\n",
"source": [ "tool_use_agent = AgentId(\"tool_use_agent\", \"default\")\n",
"# Start processing messages.\n", "response = await runtime.send_message(Message(\"What is the stock price of NVDA on 2024/06/01?\"), tool_use_agent)\n",
"runtime.start()\n", "print(response.content)\n",
"# Send a direct message to the tool agent.\n", "# Stop processing messages.\n",
"tool_use_agent = AgentId(\"tool-use-agent\", \"default\")\n", "await runtime.stop()"
"response = await runtime.send_message(Message(\"What is the stock price of NVDA on 2024/06/01?\"), tool_use_agent)\n", ]
"print(response.content)\n", },
"# Stop processing messages.\n", {
"await runtime.stop()" "cell_type": "markdown",
] "metadata": {},
}, "source": [
{ "See [samples](https://github.com/microsoft/agnext/tree/main/python/samples#tool-use-examples)\n",
"cell_type": "markdown", "for more examples of using tools with agents, including how to use\n",
"metadata": {}, "broadcast communication model for tool execution, and how to intercept tool\n",
"source": [ "execution for human-in-the-loop approval."
"See [samples](https://github.com/microsoft/agnext/tree/main/python/samples#tool-use-examples)\n", ]
"for more examples of using tools with agents, including how to use\n", }
"broadcast communication model for tool execution, and how to intercept tool\n", ],
"execution for human-in-the-loop approval." "metadata": {
] "kernelspec": {
} "display_name": "agnext",
], "language": "python",
"metadata": { "name": "python3"
"kernelspec": { },
"display_name": "agnext", "language_info": {
"language": "python", "codemirror_mode": {
"name": "python3" "name": "ipython",
}, "version": 3
"language_info": { },
"codemirror_mode": { "file_extension": ".py",
"name": "ipython", "mimetype": "text/x-python",
"version": 3 "name": "python",
}, "nbconvert_exporter": "python",
"file_extension": ".py", "pygments_lexer": "ipython3",
"mimetype": "text/x-python", "version": "3.11.9"
"name": "python", }
"nbconvert_exporter": "python", },
"pygments_lexer": "ipython3", "nbformat": 4,
"version": "3.11.9" "nbformat_minor": 2
}
},
"nbformat": 4,
"nbformat_minor": 2
} }

2
python/docs/src/index.rst
View File

@ -30,6 +30,8 @@ To learn about the core concepts of AGNext, read the `overview <core-concepts/ov
getting-started/model-clients getting-started/model-clients
getting-started/tools getting-started/tools
getting-started/multi-agent-design-patterns getting-started/multi-agent-design-patterns
getting-started/group-chat
getting-started/reflection
.. toctree:: .. toctree::
:caption: Guides :caption: Guides

2
python/src/agnext/components/tool_agent/__init__.py
View File

@ -1,3 +1,4 @@
from ._caller_loop import tool_agent_caller_loop
from ._tool_agent import ( from ._tool_agent import (
InvalidToolArgumentsException, InvalidToolArgumentsException,
ToolAgent, ToolAgent,
@ -12,4 +13,5 @@ __all__ = [
"ToolNotFoundException", "ToolNotFoundException",
"InvalidToolArgumentsException", "InvalidToolArgumentsException",
"ToolExecutionException", "ToolExecutionException",
"tool_agent_caller_loop",
] ]

77
python/src/agnext/components/tool_agent/_caller_loop.py Normal file
View File

@ -0,0 +1,77 @@
import asyncio
from typing import List
from ...components import FunctionCall
from ...core import AgentId, AgentRuntime, BaseAgent, CancellationToken
from ..models import (
AssistantMessage,
ChatCompletionClient,
FunctionExecutionResult,
FunctionExecutionResultMessage,
LLMMessage,
)
from ..tools import Tool, ToolSchema
from ._tool_agent import ToolException
async def tool_agent_caller_loop(
caller: BaseAgent | AgentRuntime,
tool_agent_id: AgentId,
model_client: ChatCompletionClient,
input_messages: List[LLMMessage],
tool_schema: List[ToolSchema] | List[Tool],
cancellation_token: CancellationToken | None = None,
caller_source: str = "assistant",
) -> List[LLMMessage]:
"""Start a caller loop for a tool agent. This function sends messages to the tool agent
and the model client in an alternating fashion until the model client stops generating tool calls.
Args:
tool_agent_id (AgentId): The Agent ID of the tool agent.
input_messages (List[LLMMessage]): The list of input messages.
model_client (ChatCompletionClient): The model client to use for the model API.
tool_schema (List[Tool | ToolSchema]): The list of tools that the model can use.
Returns:
List[LLMMessage]: The list of output messages created in the caller loop.
"""
generated_messages: List[LLMMessage] = []
# Get a response from the model.
response = await model_client.create(input_messages, tools=tool_schema, cancellation_token=cancellation_token)
# Add the response to the generated messages.
generated_messages.append(AssistantMessage(content=response.content, source=caller_source))
# Keep iterating until the model stops generating tool calls.
while isinstance(response.content, list) and all(isinstance(item, FunctionCall) for item in response.content):
# Execute functions called by the model by sending messages to tool agent.
results: List[FunctionExecutionResult | BaseException] = await asyncio.gather(
*[
caller.send_message(
message=call,
recipient=tool_agent_id,
cancellation_token=cancellation_token,
)
for call in response.content
],
return_exceptions=True,
)
# Combine the results into a single response and handle exceptions.
function_results: List[FunctionExecutionResult] = []
for result in results:
if isinstance(result, FunctionExecutionResult):
function_results.append(result)
elif isinstance(result, ToolException):
function_results.append(FunctionExecutionResult(content=f"Error: {result}", call_id=result.call_id))
elif isinstance(result, BaseException):
raise result # Unexpected exception.
generated_messages.append(FunctionExecutionResultMessage(content=function_results))
# Query the model again with the new response.
response = await model_client.create(
input_messages + generated_messages, tools=tool_schema, cancellation_token=cancellation_token
)
generated_messages.append(AssistantMessage(content=response.content, source=caller_source))
# Return the generated messages.
return generated_messages

105
python/tests/test_tool_agent.py
View File

@ -1,19 +1,34 @@
import asyncio import asyncio
import json import json
from typing import Any, AsyncGenerator, List
import pytest import pytest
from openai.resources.chat.completions import AsyncCompletions
from openai.types.chat.chat_completion import ChatCompletion, Choice
from openai.types.chat.chat_completion_chunk import ChatCompletionChunk
from openai.types.chat.chat_completion_message import ChatCompletionMessage
from openai.types.chat.chat_completion_message_tool_call import ChatCompletionMessageToolCall, Function
from openai.types.completion_usage import CompletionUsage
from agnext.application import SingleThreadedAgentRuntime from agnext.application import SingleThreadedAgentRuntime
from agnext.components import FunctionCall from agnext.components import FunctionCall
from agnext.components.models import FunctionExecutionResult
from agnext.components.tool_agent import ( from agnext.components.tool_agent import (
InvalidToolArgumentsException, InvalidToolArgumentsException,
ToolAgent, ToolAgent,
ToolExecutionException, ToolExecutionException,
ToolNotFoundException, ToolNotFoundException,
tool_agent_caller_loop,
)
from agnext.components.tools import FunctionTool, Tool
from agnext.core import CancellationToken, AgentId
from agnext.components.models import (
AssistantMessage,
FunctionExecutionResult,
FunctionExecutionResultMessage,
OpenAIChatCompletionClient,
UserMessage,
) )
from agnext.components.tools import FunctionTool from agnext.components.tools import FunctionTool
from agnext.core import CancellationToken
from agnext.core import AgentId
def _pass_function(input: str) -> str: def _pass_function(input: str) -> str:
@ -29,6 +44,60 @@ async def _async_sleep_function(input: str) -> str:
return "pass" return "pass"
class _MockChatCompletion:
def __init__(self, model: str = "gpt-4o") -> None:
self._saved_chat_completions: List[ChatCompletion] = [
ChatCompletion(
id="id1",
choices=[
Choice(
finish_reason="tool_calls",
index=0,
message=ChatCompletionMessage(
content=None,
tool_calls=[
ChatCompletionMessageToolCall(
id="1",
type="function",
function=Function(
name="pass",
arguments=json.dumps({"input": "pass"}),
),
)
],
role="assistant",
),
)
],
created=0,
model=model,
object="chat.completion",
usage=CompletionUsage(prompt_tokens=0, completion_tokens=0, total_tokens=0),
),
ChatCompletion(
id="id2",
choices=[
Choice(
finish_reason="stop", index=0, message=ChatCompletionMessage(content="Hello", role="assistant")
)
],
created=0,
model=model,
object="chat.completion",
usage=CompletionUsage(prompt_tokens=0, completion_tokens=0, total_tokens=0),
),
]
self._curr_index = 0
async def mock_create(
self, *args: Any, **kwargs: Any
) -> ChatCompletion | AsyncGenerator[ChatCompletionChunk, None]:
await asyncio.sleep(0.1)
completion = self._saved_chat_completions[self._curr_index]
self._curr_index += 1
return completion
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_tool_agent() -> None: async def test_tool_agent() -> None:
runtime = SingleThreadedAgentRuntime() runtime = SingleThreadedAgentRuntime()
@ -74,3 +143,33 @@ async def test_tool_agent() -> None:
await result_future await result_future
await runtime.stop() await runtime.stop()
@pytest.mark.asyncio
async def test_caller_loop(monkeypatch: pytest.MonkeyPatch) -> None:
mock = _MockChatCompletion(model="gpt-4o-2024-05-13")
monkeypatch.setattr(AsyncCompletions, "create", mock.mock_create)
client = OpenAIChatCompletionClient(model="gpt-4o-2024-05-13", api_key="api_key")
tools : List[Tool] = [FunctionTool(_pass_function, name="pass", description="Pass function")]
runtime = SingleThreadedAgentRuntime()
await runtime.register(
"tool_agent",
lambda: ToolAgent(
description="Tool agent",
tools=tools,
),
)
agent = AgentId("tool_agent", "default")
runtime.start()
messages = await tool_agent_caller_loop(
runtime,
agent,
client,
[UserMessage(content="Hello", source="user")],
tool_schema=tools
)
assert len(messages) == 3
assert isinstance(messages[0], AssistantMessage)
assert isinstance(messages[1], FunctionExecutionResultMessage)
assert isinstance(messages[2], AssistantMessage)
await runtime.stop()