TypeError: Cannot instantiate typing.Union #1411
Description
Below is my full code
from __future__ import annotations
import asyncio
from typing import Optional
from tavily import TavilyClient
from agents import Agent, Runner, function_tool, set_tracing_disabled
from agents.extensions.models.litellm_model import LitellmModel
from agents.extensions.visualization import draw_graph
from openai import AsyncOpenAI
from openai.types.responses import ResponseTextDeltaEvent
from google.colab import userdata
# Get API keys
GEMINI_API_KEY = userdata.get('GOOGLE_API_KEY')
TAVILY_API_KEY = "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" # Consider moving to userdata
# Model configuration
MODEL = 'gemini/gemini-2.0-flash'
# Initialize Tavily client for research
tavily_client = TavilyClient(TAVILY_API_KEY)
# Disable tracing for performance
set_tracing_disabled(disabled=True)
@function_tool
def send_email(body: str) -> str:
"""Send an email with the specified body text."""
print(f"Sending email with body: {body[:100]}...") # Show first 100 chars
return "The email has been sent successfully"
@function_tool
def research(query: str) -> list:
"""Perform web research on a given query."""
print(f"Researching: {query}")
response = tavily_client.search(query=query)
return response["results"]
# Email Agent
email_agent = Agent(
name="Email Agent",
instructions="""
You are the Email Agent.
Your sole responsibility is to send emails using the `send_email` tool.
The `send_email` tool has a single parameter: `body` (string), which is the full text of the email.
When you are asked to send an email:
1. Extract the exact email text from the user's request (do not add or remove words unless explicitly instructed).
2. Pass the complete email text as the `body` parameter to the `send_email` tool.
Do not answer general questions, perform research, or do any other task.
If the request is not about sending an email, hand it off to the appropriate agent.
""",
model=LitellmModel(
model=MODEL,
api_key=GEMINI_API_KEY
),
tools=[send_email],
handoff_description="Handles email sending requests."
)
# Research Agent
research_agent = Agent(
name="Research Agent",
instructions="""
You are the Research Agent.
Your job is to gather detailed, factual, and up-to-date information using the `research` tool.
The `research` tool has one required parameter: `query` (string), which is the exact search query.
For every user request you receive:
1. Identify the core research topic from the request.
2. Call the `research` tool with the topic as the `query` parameter (worded naturally for best search results).
3. Return the research results to the user in a well-structured and concise way.
Do not make up answers — always use the `research` tool for factual queries.
If the request is unrelated to research, hand it off to the correct agent.
""",
model=LitellmModel(
model=MODEL,
api_key=GEMINI_API_KEY
),
tools=[research],
handoff_description="Handles research and fact-finding tasks."
)
async def main():
"""Main function to run the agent system."""
# General Assistant (Orchestrator)
general_assistant = Agent(
name="General Assistant",
instructions="""
You are the General Assistant.
Your goal is to help the user by either:
- Answering the question directly if you have sufficient knowledge.
- Handing off the request to the correct specialized agent when it matches their role.
The specialized agents are:
1. Research Agent — for fact-checking, gathering latest news, or answering knowledge-based questions.
2. Email Agent — for composing and sending emails.
Rules:
- If the query is about research or factual updates, hand it off to the Research Agent.
- If the query is about sending an email, hand it off to the Email Agent.
- If the query involves both research and email sending, handle each part with the correct agent separately.
- Do not ask the user if they want a transfer — just hand it off automatically.
- Get back the results from the agent and return them in proper markdown format.
""",
model=LitellmModel(
model=MODEL,
api_key=GEMINI_API_KEY
),
tools=[
research_agent.as_tool(
tool_name="research_agent",
tool_description="This tool performs web searches and returns research results"
),
email_agent.as_tool(
tool_name="email_agent",
tool_description="This tool is used to send emails"
)
]
)
# Example query with both research and email tasks
query = "Give me a detailed 300-word news article on OpenAI's latest model, and send an email saying 'Tomorrow is the meeting at 10 PM, be there on time.'"
# Run the agent with streaming
result = Runner.run_streamed(general_assistant, query)
# Stream the response
async for event in result.stream_events():
if event.type == "raw_response_event" and isinstance(event.data, ResponseTextDeltaEvent):
print(event.data.delta, end="", flush=True)
print("\n") # New line after streaming completes
# Run the async main function
if __name__ == "__main__":
asyncio.run(main())ERROR :
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
[/tmp/ipython-input-3639346673.py](https://localhost:8080/#) in <cell line: 0>()
139
140
--> 141 asyncio.run(main())
17 frames
[/usr/lib/python3.11/typing.py](https://localhost:8080/#) in __call__(self, *args, **kwds)
484
485 def __call__(self, *args, **kwds):
--> 486 raise TypeError(f"Cannot instantiate {self!r}")
487
488 def __or__(self, other):
TypeError: Cannot instantiate typing.Union
----------------------------------------------------------------------------------------------------------------------
Note it was working perfectly Yesterday !
