Files
humanlayer/examples/langchain/07-grocery-slack-blocks.py
2025-01-19 18:17:50 -08:00

104 lines
2.6 KiB
Python

from datetime import datetime
from typing import Any
import langchain_core.tools as langchain_tools
from dotenv import load_dotenv
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_tool_calling_agent
from pydantic import BaseModel
from channels import (
dm_with_ceo,
)
from humanlayer import ContactChannel, SlackContactChannel
from humanlayer.core.approval import (
HumanLayer,
)
load_dotenv()
hl = HumanLayer(
verbose=True,
# run_id is optional -it can be used to identify the agent in approval history
run_id="langchain-grocery-slack-blocks",
)
task_prompt = """
You are the mealprep power assistant.
You are responsible for planning the meals and shopping
for a very busy and high-profile and health-conscious tech CEO.
choose a meal plan based on your best judgement,
i like tacos and sushi, but i'm open to new ideas.
I like to eat healthy, and I'm trying to lose weight.
Make the best decision and order the groceries. Don't confirm with me.
"""
channel = ContactChannel(
slack=SlackContactChannel(
channel_or_user_id="C07HR5JL15F",
experimental_slack_blocks=True,
),
)
@hl.require_approval(contact_channel=channel)
def buy_groceries(items: list[Any], total_cost: int) -> str:
"""purchase a cart of groceries. Include structured data about the quantity
and price, for example:
[
{
"item": "bananas",
"quantity": 2,
"price": 1.99
},
{
"item": "Organic 2% Milk, 3 x 64oz",
"quantity": 1,
"price": 11.99
},
{
"item": "Organic Chicken Breast, 2lbs",
"quantity": 1,
"price": 9.99
}
]
"""
return f"Items purchased, total cost: {total_cost}"
tools = [
langchain_tools.StructuredTool.from_function(buy_groceries),
langchain_tools.StructuredTool.from_function(hl.human_as_tool(contact_channel=channel)),
]
llm = ChatOpenAI(model="gpt-4o", temperature=0)
# Prompt for creating Tool Calling Agent
prompt = ChatPromptTemplate.from_messages(
[
(
"system",
"You are a helpful assistant.",
),
("placeholder", "{chat_history}"),
("human", "{input}"),
("placeholder", "{agent_scratchpad}"),
]
)
# Construct the Tool Calling Agent
agent = create_tool_calling_agent(llm, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
if __name__ == "__main__":
result = agent_executor.invoke({"input": task_prompt})
print("\n\n----------Result----------\n\n")
print(result)