GPT Builder Demo¶

Inspired by GPTs interface, presented at OpenAI Dev Day 2023. Construct an agent with natural language.

Here you can build your own agent...with another agent!

In [ ]:

%pip install llama-index-agent-openai
%pip install llama-index-embeddings-openai
%pip install llama-index-llms-openai

%pip install llama-index-agent-openai %pip install llama-index-embeddings-openai %pip install llama-index-llms-openai

In [ ]:

import os

os.environ["OPENAI_API_KEY"] = "sk-..."

import os os.environ["OPENAI_API_KEY"] = "sk-..."

In [ ]:

from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.llms.openai import OpenAI
from llama_index.core import Settings

llm = OpenAI(model="gpt-4")
Settings.llm = llm
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")

from llama_index.embeddings.openai import OpenAIEmbedding from llama_index.llms.openai import OpenAI from llama_index.core import Settings llm = OpenAI(model="gpt-4") Settings.llm = llm Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")

Define Candidate Tools¶

We also define a tool retriever to retrieve candidate tools.

In this setting we define tools as different Wikipedia pages.

In [ ]:

from llama_index.core import SimpleDirectoryReader

from llama_index.core import SimpleDirectoryReader

In [ ]:

wiki_titles = ["Toronto", "Seattle", "Chicago", "Boston", "Houston"]

wiki_titles = ["Toronto", "Seattle", "Chicago", "Boston", "Houston"]

In [ ]:

from pathlib import Path

import requests

for title in wiki_titles:
    response = requests.get(
        "https://en.wikipedia.org/w/api.php",
        params={
            "action": "query",
            "format": "json",
            "titles": title,
            "prop": "extracts",
            # 'exintro': True,
            "explaintext": True,
        },
    ).json()
    page = next(iter(response["query"]["pages"].values()))
    wiki_text = page["extract"]

    data_path = Path("data")
    if not data_path.exists():
        Path.mkdir(data_path)

    with open(data_path / f"{title}.txt", "w") as fp:
        fp.write(wiki_text)

from pathlib import Path import requests for title in wiki_titles: response = requests.get( "https://en.wikipedia.org/w/api.php", params={ "action": "query", "format": "json", "titles": title, "prop": "extracts", # 'exintro': True, "explaintext": True, }, ).json() page = next(iter(response["query"]["pages"].values())) wiki_text = page["extract"] data_path = Path("data") if not data_path.exists(): Path.mkdir(data_path) with open(data_path / f"{title}.txt", "w") as fp: fp.write(wiki_text)

In [ ]:

# Load all wiki documents
city_docs = {}
for wiki_title in wiki_titles:
    city_docs[wiki_title] = SimpleDirectoryReader(
        input_files=[f"data/{wiki_title}.txt"]
    ).load_data()

# Load all wiki documents city_docs = {} for wiki_title in wiki_titles: city_docs[wiki_title] = SimpleDirectoryReader( input_files=[f"data/{wiki_title}.txt"] ).load_data()

Build Query Tool for Each Document¶

In [ ]:

from llama_index.core import VectorStoreIndex
from llama_index.agent.openai import OpenAIAgent
from llama_index.core.tools import QueryEngineTool, ToolMetadata
from llama_index.core import VectorStoreIndex

# Build tool dictionary
tool_dict = {}

for wiki_title in wiki_titles:
    # build vector index
    vector_index = VectorStoreIndex.from_documents(
        city_docs[wiki_title],
    )
    # define query engines
    vector_query_engine = vector_index.as_query_engine(llm=llm)

    # define tools
    vector_tool = QueryEngineTool(
        query_engine=vector_query_engine,
        metadata=ToolMetadata(
            name=wiki_title,
            description=("Useful for questions related to" f" {wiki_title}"),
        ),
    )
    tool_dict[wiki_title] = vector_tool

from llama_index.core import VectorStoreIndex from llama_index.agent.openai import OpenAIAgent from llama_index.core.tools import QueryEngineTool, ToolMetadata from llama_index.core import VectorStoreIndex # Build tool dictionary tool_dict = {} for wiki_title in wiki_titles: # build vector index vector_index = VectorStoreIndex.from_documents( city_docs[wiki_title], ) # define query engines vector_query_engine = vector_index.as_query_engine(llm=llm) # define tools vector_tool = QueryEngineTool( query_engine=vector_query_engine, metadata=ToolMetadata( name=wiki_title, description=("Useful for questions related to" f" {wiki_title}"), ), ) tool_dict[wiki_title] = vector_tool

Define Tool Retriever¶

In [ ]:

# define an "object" index and retriever over these tools
from llama_index.core import VectorStoreIndex
from llama_index.core.objects import ObjectIndex

tool_index = ObjectIndex.from_objects(
    list(tool_dict.values()),
    index_cls=VectorStoreIndex,
)
tool_retriever = tool_index.as_retriever(similarity_top_k=1)

# define an "object" index and retriever over these tools from llama_index.core import VectorStoreIndex from llama_index.core.objects import ObjectIndex tool_index = ObjectIndex.from_objects( list(tool_dict.values()), index_cls=VectorStoreIndex, ) tool_retriever = tool_index.as_retriever(similarity_top_k=1)

Load Data¶

Here we load wikipedia pages from different cities.

Define Meta-Tools for GPT Builder¶

In [ ]:

from llama_index.core.llms import ChatMessage
from llama_index.core import ChatPromptTemplate
from typing import List

GEN_SYS_PROMPT_STR = """\
Task information is given below. 

Given the task, please generate a system prompt for an OpenAI-powered bot to solve this task: 
{task} \
"""

gen_sys_prompt_messages = [
    ChatMessage(
        role="system",
        content="You are helping to build a system prompt for another bot.",
    ),
    ChatMessage(role="user", content=GEN_SYS_PROMPT_STR),
]

GEN_SYS_PROMPT_TMPL = ChatPromptTemplate(gen_sys_prompt_messages)


agent_cache = {}


def create_system_prompt(task: str):
    """Create system prompt for another agent given an input task."""
    llm = OpenAI(llm="gpt-4")
    fmt_messages = GEN_SYS_PROMPT_TMPL.format_messages(task=task)
    response = llm.chat(fmt_messages)
    return response.message.content


def get_tools(task: str):
    """Get the set of relevant tools to use given an input task."""
    subset_tools = tool_retriever.retrieve(task)
    return [t.metadata.name for t in subset_tools]


def create_agent(system_prompt: str, tool_names: List[str]):
    """Create an agent given a system prompt and an input set of tools."""
    llm = OpenAI(model="gpt-4")
    try:
        # get the list of tools
        input_tools = [tool_dict[tn] for tn in tool_names]

        agent = OpenAIAgent.from_tools(input_tools, llm=llm, verbose=True)
        agent_cache["agent"] = agent
        return_msg = "Agent created successfully."
    except Exception as e:
        return_msg = f"An error occurred when building an agent. Here is the error: {repr(e)}"
    return return_msg

from llama_index.core.llms import ChatMessage from llama_index.core import ChatPromptTemplate from typing import List GEN_SYS_PROMPT_STR = """\ Task information is given below. Given the task, please generate a system prompt for an OpenAI-powered bot to solve this task: {task} \ """ gen_sys_prompt_messages = [ ChatMessage( role="system", content="You are helping to build a system prompt for another bot.", ), ChatMessage(role="user", content=GEN_SYS_PROMPT_STR), ] GEN_SYS_PROMPT_TMPL = ChatPromptTemplate(gen_sys_prompt_messages) agent_cache = {} def create_system_prompt(task: str): """Create system prompt for another agent given an input task.""" llm = OpenAI(llm="gpt-4") fmt_messages = GEN_SYS_PROMPT_TMPL.format_messages(task=task) response = llm.chat(fmt_messages) return response.message.content def get_tools(task: str): """Get the set of relevant tools to use given an input task.""" subset_tools = tool_retriever.retrieve(task) return [t.metadata.name for t in subset_tools] def create_agent(system_prompt: str, tool_names: List[str]): """Create an agent given a system prompt and an input set of tools.""" llm = OpenAI(model="gpt-4") try: # get the list of tools input_tools = [tool_dict[tn] for tn in tool_names] agent = OpenAIAgent.from_tools(input_tools, llm=llm, verbose=True) agent_cache["agent"] = agent return_msg = "Agent created successfully." except Exception as e: return_msg = f"An error occurred when building an agent. Here is the error: {repr(e)}" return return_msg

In [ ]:

from llama_index.core.tools import FunctionTool

system_prompt_tool = FunctionTool.from_defaults(fn=create_system_prompt)
get_tools_tool = FunctionTool.from_defaults(fn=get_tools)
create_agent_tool = FunctionTool.from_defaults(fn=create_agent)

from llama_index.core.tools import FunctionTool system_prompt_tool = FunctionTool.from_defaults(fn=create_system_prompt) get_tools_tool = FunctionTool.from_defaults(fn=get_tools) create_agent_tool = FunctionTool.from_defaults(fn=create_agent)

In [ ]:

GPT_BUILDER_SYS_STR = """\
You are helping to construct an agent given a user-specified task. You should generally use the tools in this order to build the agent.

1) Create system prompt tool: to create the system prompt for the agent.
2) Get tools tool: to fetch the candidate set of tools to use.
3) Create agent tool: to create the final agent.
"""

prefix_msgs = [ChatMessage(role="system", content=GPT_BUILDER_SYS_STR)]


builder_agent = OpenAIAgent.from_tools(
    tools=[system_prompt_tool, get_tools_tool, create_agent_tool],
    prefix_messages=prefix_msgs,
    verbose=True,
)

GPT_BUILDER_SYS_STR = """\ You are helping to construct an agent given a user-specified task. You should generally use the tools in this order to build the agent. 1) Create system prompt tool: to create the system prompt for the agent. 2) Get tools tool: to fetch the candidate set of tools to use. 3) Create agent tool: to create the final agent. """ prefix_msgs = [ChatMessage(role="system", content=GPT_BUILDER_SYS_STR)] builder_agent = OpenAIAgent.from_tools( tools=[system_prompt_tool, get_tools_tool, create_agent_tool], prefix_messages=prefix_msgs, verbose=True, )

In [ ]:

builder_agent.query("Build an agent that can tell me about Toronto.")

builder_agent.query("Build an agent that can tell me about Toronto.")

Added user message to memory: Build an agent that can tell me about Toronto.
=== Calling Function ===
Calling function: create_system_prompt with args: {
  "task": "tell me about Toronto"
}
Got output: "Generate a brief summary about Toronto, including its history, culture, landmarks, and notable features."
========================

=== Calling Function ===
Calling function: get_tools with args: {
  "task": "tell me about Toronto"
}
Got output: ['Toronto']
========================

=== Calling Function ===
Calling function: create_agent with args: {
  "system_prompt": "Generate a brief summary about Toronto, including its history, culture, landmarks, and notable features.",
  "tool_names": ["Toronto"]
}
Got output: Agent created successfully.
========================

Out[ ]:

Response(response='The agent has been successfully created. It can now provide information about Toronto, including its history, culture, landmarks, and notable features.', source_nodes=[], metadata=None)

In [ ]:

city_agent = agent_cache["agent"]

city_agent = agent_cache["agent"]

In [ ]:

response = city_agent.query("Tell me about the parks in Toronto")
print(str(response))

response = city_agent.query("Tell me about the parks in Toronto") print(str(response))

Added user message to memory: Tell me about the parks in Toronto
Toronto is known for its beautiful and diverse parks. Here are a few of the most popular ones:

1. **High Park**: This is Toronto's largest public park featuring many hiking trails, sports facilities, a beautiful lakefront, convenient parking, easy public transit access, a dog park, a zoo, and playgrounds for children. It's also known for its spring cherry blossoms.

2. **Toronto Islands**: A group of small islands located just off the shore of the city's downtown district, offering stunning views of the city skyline. The islands provide a great escape from the city with their car-free environment, picnic spots, swimming beaches, and Centreville Amusement Park.

3. **Trinity Bellwoods Park**: A popular park in the downtown area, it's a great place for picnics, sports, dog-walking, or just relaxing. It also has a community recreation centre with a pool and gym.

4. **Rouge National Urban Park**: Located in the city's east end, this is Canada's first national urban park. It offers hiking, swimming, camping, and a chance to learn about the area's cultural and agricultural heritage.

5. **Riverdale Farm**: This 7.5-acre farm in the heart of Toronto provides an opportunity to experience farm life and interact with a variety of farm animals.

6. **Evergreen Brick Works**: A former industrial site that has been transformed into an eco-friendly community center with a park, farmers market, and cultural events.

7. **Scarborough Bluffs Park**: Offers a unique natural environment with stunning views of Lake Ontario from atop the bluffs.

8. **Edwards Gardens**: A beautiful botanical garden located in North York, perfect for a peaceful walk surrounded by nature.

9. **Sunnybrook Park**: A large public park that offers many recreational activities including horseback riding, sports fields, and picnic areas.

10. **Cherry Beach**: Located on the waterfront, this park offers a sandy beach, picnic areas, and a dog off-leash area. It's a great spot for swimming, sunbathing, and barbecuing.

These parks offer a variety of experiences, from urban amenities to natural beauty, making Toronto a great city for outdoor enthusiasts.