Examples
- Examples by Category
- Examples by Notebook
- Notebooks
- Using RetrieveChat Powered by MongoDB Atlas for Retrieve Augmented Code Generation and Question Answering
- Using RetrieveChat Powered by PGVector for Retrieve Augmented Code Generation and Question Answering
- Using RetrieveChat with Qdrant for Retrieve Augmented Code Generation and Question Answering
- Agent Tracking with AgentOps
- AgentOptimizer: An Agentic Way to Train Your LLM Agent
- Task Solving with Code Generation, Execution and Debugging
- Assistants with Azure Cognitive Search and Azure Identity
- CaptainAgent
- Usage tracking with AutoGen
- Agent Chat with custom model loading
- Agent Chat with Multimodal Models: DALLE and GPT-4V
- Use AutoGen in Databricks with DBRX
- Auto Generated Agent Chat: Task Solving with Provided Tools as Functions
- Task Solving with Provided Tools as Functions (Asynchronous Function Calls)
- Writing a software application using function calls
- Currency Calculator: Task Solving with Provided Tools as Functions
- Groupchat with Llamaindex agents
- Group Chat
- Group Chat with Retrieval Augmented Generation
- Group Chat with Customized Speaker Selection Method
- FSM - User can input speaker transition constraints
- Perform Research with Multi-Agent Group Chat
- StateFlow: Build Workflows through State-Oriented Actions
- Group Chat with Coder and Visualization Critic
- Using Guidance with AutoGen
- Auto Generated Agent Chat: Task Solving with Code Generation, Execution, Debugging & Human Feedback
- Generate Dalle Images With Conversable Agents
- Auto Generated Agent Chat: Function Inception
- Auto Generated Agent Chat: Task Solving with Langchain Provided Tools as Functions
- Engaging with Multimodal Models: GPT-4V in AutoGen
- Agent Chat with Multimodal Models: LLaVA
- Runtime Logging with AutoGen
- Agent with memory using Mem0
- Solving Multiple Tasks in a Sequence of Async Chats
- Solving Multiple Tasks in a Sequence of Chats
- Nested Chats for Tool Use in Conversational Chess
- Conversational Chess using non-OpenAI clients
- Solving Complex Tasks with A Sequence of Nested Chats
- Solving Complex Tasks with Nested Chats
- OptiGuide with Nested Chats in AutoGen
- Chat with OpenAI Assistant using function call in AutoGen: OSS Insights for Advanced GitHub Data Analysis
- Auto Generated Agent Chat: Group Chat with GPTAssistantAgent
- RAG OpenAI Assistants in AutoGen
- OpenAI Assistants in AutoGen
- Auto Generated Agent Chat: GPTAssistant with Code Interpreter
- Agent Observability with OpenLIT
- Auto Generated Agent Chat: Collaborative Task Solving with Coding and Planning Agent
- ReasoningAgent - Advanced LLM Reasoning with Multiple Search Strategies
- SocietyOfMindAgent
- SQL Agent for Spider text-to-SQL benchmark
- Interactive LLM Agent Dealing with Data Stream
- Structured output
- WebSurferAgent
- Swarm Orchestration with AG2
- Using a local Telemetry server to monitor a GraphRAG agent
- Trip planning with a FalkorDB GraphRAG agent using a Swarm
- (Legacy) Implement Swarm-style orchestration with GroupChat
- Chatting with a teachable agent
- Making OpenAI Assistants Teachable
- Auto Generated Agent Chat: Teaching AI New Skills via Natural Language Interaction
- Preprocessing Chat History with `TransformMessages`
- Auto Generated Agent Chat: Collaborative Task Solving with Multiple Agents and Human Users
- Translating Video audio using Whisper and GPT-3.5-turbo
- Auto Generated Agent Chat: Solving Tasks Requiring Web Info
- Web Scraping using Apify Tools
- Websockets: Streaming input and output using websockets
- Solving Multiple Tasks in a Sequence of Chats with Different Conversable Agent Pairs
- Demonstrating the `AgentEval` framework using the task of solving math problems as an example
- Agent Chat with Async Human Inputs
- Automatically Build Multi-agent System from Agent Library
- AutoBuild
- A Uniform interface to call different LLMs
- From Dad Jokes To Sad Jokes: Function Calling with GPTAssistantAgent
- Language Agent Tree Search
- Mitigating Prompt hacking with JSON Mode in Autogen
- Using RetrieveChat for Retrieve Augmented Code Generation and Question Answering
- Using Neo4j's graph database with AG2 agents for Question & Answering
- Enhanced Swarm Orchestration with AG2
- Cross-Framework LLM Tool Integration with AG2
- RealtimeAgent in a Swarm Orchestration
- ReasoningAgent - Advanced LLM Reasoning with Multiple Search Strategies
- Application Gallery
From Dad Jokes To Sad Jokes: Function Calling with GPTAssistantAgent
Autogen allows GPTAssistantAgent
to be augmented with “tools” —
pre-defined functions or capabilities — that extend its ability to
handle specific tasks, similar to how one might natively utilize tools
in the OpenAI Assistant’s
API.
In this notebook, we create a basic Multi-Agent System using Autogen’s
GPTAssistantAgent
to convert Dad jokes on a specific topic into Sad
jokes. It consists of a “Dad” agent which has the ability to search the
Dad Joke API and a “Sad Joker” agent
which converts the Dad jokes into Sad jokes. The Sad Joker then writes
the sad jokes into a txt file.
In this process we demonstrate how to call tools and perform function
calling for GPTAssistantAgent
.
Requirements
AutoGen requires Python 3.9 or newer. For this notebook, please install
autogen
:
%pip install autogen
Import Dependencies
from typing import Annotated, Literal
import requests
import autogen
from autogen import UserProxyAgent
from autogen.agentchat.contrib.gpt_assistant_agent import GPTAssistantAgent
from autogen.function_utils import get_function_schema
config_list = autogen.config_list_from_json(
env_or_file="OAI_CONFIG_LIST",
)
Creating the Functions
We need to create functions for our Agents to call.
This function calls the Dad Joke API with a search term that the agent creates and returns a list of dad jokes.
def get_dad_jokes(search_term: str, page: int = 1, limit: int = 10) -> str:
"""
Fetches a list of dad jokes based on a search term.
Parameters:
- search_term: The search term to find jokes about.
- page: The page number of results to fetch (default is 1).
- limit: The number of results to return per page (default is 20, max is 30).
Returns:
A list of dad jokes.
"""
url = "https://icanhazdadjoke.com/search"
headers = {"Accept": "application/json"}
params = {"term": search_term, "page": page, "limit": limit}
response = requests.get(url, headers=headers, params=params)
if response.status_code == 200:
data = response.json()
jokes = [joke["joke"] for joke in data["results"]]
return jokes
else:
return f"Failed to fetch jokes, status code: {response.status_code}"
# Example Dad Jokes Function Usage:
jokes = get_dad_jokes("cats")
print(jokes)
['Where do cats write notes?\r\nScratch Paper!', 'It was raining cats and dogs the other day. I almost stepped in a poodle.', 'What do you call a group of disorganized cats? A cat-tastrophe.', 'I accidentally took my cats meds last night. Don’t ask meow.', 'What do you call a pile of cats? A Meowtain.', 'Animal Fact #25: Most bobcats are not named bob.']
This function allows the Agents to write to a txt file.
def write_to_txt(content: str, filename: str = "dad_jokes.txt"):
"""
Writes a formatted string to a text file.
Parameters:
- content: The formatted string to write.
- filename: The name of the file to write to. Defaults to "output.txt".
"""
with open(filename, "w") as file:
file.write(content)
# Example Write to TXT Function Usage:
content = "\n".join(jokes) # Format the jokes from the above example
write_to_txt(content)
Create Function Schemas
In order to use the functions within our GPTAssistantAgents, we need to
generate function schemas. This can be done by using
get_function_schema
# Assistant API Tool Schema for get_dad_jokes
get_dad_jokes_schema = get_function_schema(
get_dad_jokes,
name="get_dad_jokes",
description="Fetches a list of dad jokes based on a search term. Allows pagination with page and limit parameters.",
)
# Assistant API Tool Schema for write_to_txt
write_to_txt_schema = get_function_schema(
write_to_txt,
name="write_to_txt",
description="Writes a formatted string to a text file. If the file does not exist, it will be created. If the file does exist, it will be overwritten.",
)
The return type of the function 'write_to_txt' is not annotated. Although annotating it is optional, the function should return either a string, a subclass of 'pydantic.BaseModel'.
Creating the Agents
In this section we create and configure our Dad and Sad Joker Agents
Set up the User Proxy
user_proxy = UserProxyAgent(
name="user_proxy",
is_termination_msg=lambda msg: "TERMINATE" in msg["content"],
human_input_mode="NEVER",
max_consecutive_auto_reply=1,
)
The Dad Agent
We create the Dad agent using GPTAssistantAgent
, in order for us to
enable the Dad to use the get_dad_jokes
function we need to provide it
the function’s specification in our llm_config
.
We format the tools
within our llm_config
in the same format as
provided in the OpenAI Assistant tools
docs.
the_dad = GPTAssistantAgent(
name="the_dad",
instructions="""
As 'The Dad', your primary role is to entertain by fetching dad jokes which the sad joker will transform into 'sad jokes' based on a given theme. When provided with a theme, such as 'plants' or 'animals', your task is as follows:
1. Use the 'get_dad_jokes' function to search for dad jokes related to the provided theme by providing a search term related to the theme. Fetch a list of jokes that are relevant to the theme.
2. Present these jokes to the sad joker in a format that is clear and easy to read, preparing them for transformation.
Remember, the team's goal is to creatively adapt the essence of each dad joke to fit the 'sad joke' format, all while staying true to the theme provided by the user.
""",
overwrite_instructions=True, # overwrite any existing instructions with the ones provided
overwrite_tools=True, # overwrite any existing tools with the ones provided
llm_config={
"config_list": config_list,
"tools": [get_dad_jokes_schema],
},
)
OpenAI client config of GPTAssistantAgent(the_dad) - model: gpt-4-1106-preview
Matching assistant found, using the first matching assistant: {'id': 'asst_BLBUwYPugb1UR2jQMGAA7RtU', 'created_at': 1714660644, 'description': None, 'file_ids': [], 'instructions': "\n As 'The Dad', your primary role is to entertain by fetching dad jokes which the sad joker will transform into 'sad jokes' based on a given theme. When provided with a theme, such as 'plants' or 'animals', your task is as follows:\n\n 1. Use the 'get_dad_jokes' function to search for dad jokes related to the provided theme by providing a search term related to the theme. Fetch a list of jokes that are relevant to the theme.\n 2. Present these jokes to the sad joker in a format that is clear and easy to read, preparing them for transformation.\n\n Remember, the team's goal is to creatively adapt the essence of each dad joke to fit the 'sad joke' format, all while staying true to the theme provided by the user.\n ", 'metadata': {}, 'model': 'gpt-4-1106-preview', 'name': 'the_dad', 'object': 'assistant', 'tools': [ToolFunction(function=FunctionDefinition(name='get_dad_jokes', description='Fetches a list of dad jokes based on a search term. Allows pagination with page and limit parameters.', parameters={'type': 'object', 'properties': {'search_term': {'type': 'string', 'description': 'search_term'}, 'page': {'type': 'integer', 'default': 1, 'description': 'page'}, 'limit': {'type': 'integer', 'default': 10, 'description': 'limit'}}, 'required': ['search_term']}), type='function')]}
Next, we register the get_dad_jokes
function with the Dad
GPTAssistantAgent
# Register get_dad_jokes with the_dad GPTAssistantAgent
the_dad.register_function(
function_map={
"get_dad_jokes": get_dad_jokes,
},
)
The Sad Joker Agent
We then create and configure the Sad Joker agent in a similar manner to the Dad agent above.
the_sad_joker = GPTAssistantAgent(
name="the_sad_joker",
instructions="""
As 'The Sad Joker', your unique role is to take dad jokes and creatively transform them into 'sad jokes'. When you receive a list of dad jokes, themed around topics like 'plants' or 'animals', you should:
1. Read through each dad joke carefully, understanding its theme and punchline.
2. Creatively alter the joke to change its mood from humorous to somber or melancholic. This may involve tweaking the punchline, modifying the setup, or even completely reimagining the joke while keeping it relevant to the original theme.
3. Ensure your transformations maintain a clear connection to the original theme and are understandable as adaptations of the dad jokes provided.
4. Write your transformed sad jokes to a text file using the 'write_to_txt' function. Use meaningful file names that reflect the theme or the nature of the jokes within, unless a specific filename is requested.
Your goal is not just to alter the mood of the jokes but to do so in a way that is creative, thoughtful, and respects the essence of the original humor. Remember, while the themes might be light-hearted, your transformations should offer a melancholic twist that makes them uniquely 'sad jokes'.
""",
overwrite_instructions=True, # overwrite any existing instructions with the ones provided
overwrite_tools=True, # overwrite any existing tools with the ones provided
llm_config={
"config_list": config_list,
"tools": [write_to_txt_schema],
},
)
OpenAI client config of GPTAssistantAgent(the_sad_joker) - model: gpt-4-1106-preview
Matching assistant found, using the first matching assistant: {'id': 'asst_HzB75gkobafXZhkuIAmiBiai', 'created_at': 1714660668, 'description': None, 'file_ids': [], 'instructions': "\n As 'The Sad Joker', your unique role is to take dad jokes and creatively transform them into 'sad jokes'. When you receive a list of dad jokes, themed around topics like 'plants' or 'animals', you should:\n\n 1. Read through each dad joke carefully, understanding its theme and punchline.\n 2. Creatively alter the joke to change its mood from humorous to somber or melancholic. This may involve tweaking the punchline, modifying the setup, or even completely reimagining the joke while keeping it relevant to the original theme.\n 3. Ensure your transformations maintain a clear connection to the original theme and are understandable as adaptations of the dad jokes provided.\n 4. Write your transformed sad jokes to a text file using the 'write_to_txt' function. Use meaningful file names that reflect the theme or the nature of the jokes within, unless a specific filename is requested.\n\n Your goal is not just to alter the mood of the jokes but to do so in a way that is creative, thoughtful, and respects the essence of the original humor. Remember, while the themes might be light-hearted, your transformations should offer a melancholic twist that makes them uniquely 'sad jokes'.\n ", 'metadata': {}, 'model': 'gpt-4-1106-preview', 'name': 'the_sad_joker', 'object': 'assistant', 'tools': [ToolFunction(function=FunctionDefinition(name='write_to_txt', description='Writes a formatted string to a text file. If the file does not exist, it will be created. If the file does exist, it will be overwritten.', parameters={'type': 'object', 'properties': {'content': {'type': 'string', 'description': 'content'}, 'filename': {'type': 'string', 'default': 'dad_jokes.txt', 'description': 'filename'}}, 'required': ['content']}), type='function')]}
Register the write_to_txt
function with the Sad Joker
GPTAssistantAgent
# Register get_dad_jokes with the_dad GPTAssistantAgent
the_sad_joker.register_function(
function_map={
"write_to_txt": write_to_txt,
},
)
Creating the Groupchat and Starting the Conversation
Create the groupchat
groupchat = autogen.GroupChat(agents=[user_proxy, the_dad, the_sad_joker], messages=[], max_round=15)
group_chat_manager = autogen.GroupChatManager(groupchat=groupchat, llm_config={"config_list": config_list})
Start the Conversation
user_proxy.initiate_chat(group_chat_manager, message="Jokes about cats")
user_proxy (to chat_manager):
Jokes about cats
--------------------------------------------------------------------------------
>>>>>>>> EXECUTING FUNCTION get_dad_jokes...
the_dad (to chat_manager):
Here are some cat-themed dad jokes for the sad joker to transform:
1. Where do cats write notes? Scratch Paper!
2. It was raining cats and dogs the other day. I almost stepped in a poodle.
3. What do you call a group of disorganized cats? A cat-tastrophe.
4. I accidentally took my cat's meds last night. Don’t ask meow.
5. What do you call a pile of cats? A Meowtain.
6. Animal Fact #25: Most bobcats are not named Bob.
--------------------------------------------------------------------------------
>>>>>>>> EXECUTING FUNCTION write_to_txt...
the_sad_joker (to chat_manager):
The cat-themed sad jokes have been transformed and saved to a text file named "sad_cat_jokes.txt".
--------------------------------------------------------------------------------
user_proxy (to chat_manager):
--------------------------------------------------------------------------------
ChatResult(chat_id=None, chat_history=[{'content': 'Jokes about cats', 'role': 'assistant'}, {'content': "Here are some cat-themed dad jokes for the sad joker to transform:\n\n1. Where do cats write notes? Scratch Paper!\n2. It was raining cats and dogs the other day. I almost stepped in a poodle.\n3. What do you call a group of disorganized cats? A cat-tastrophe.\n4. I accidentally took my cat's meds last night. Don’t ask meow.\n5. What do you call a pile of cats? A Meowtain.\n6. Animal Fact #25: Most bobcats are not named Bob.\n", 'name': 'the_dad', 'role': 'user'}, {'content': 'The cat-themed sad jokes have been transformed and saved to a text file named "sad_cat_jokes.txt".\n', 'name': 'the_sad_joker', 'role': 'user'}, {'content': '', 'role': 'assistant'}], summary='', cost=({'total_cost': 0.0278, 'gpt-4-1106-preview': {'cost': 0.0278, 'prompt_tokens': 2744, 'completion_tokens': 12, 'total_tokens': 2756}}, {'total_cost': 0.02194, 'gpt-4-1106-preview': {'cost': 0.02194, 'prompt_tokens': 2167, 'completion_tokens': 9, 'total_tokens': 2176}}), human_input=[])