Termination

Termination conditions in AgentOpera provide flexible mechanisms to control when a conversation should end, whether it's because a goal has been reached, a certain limit has been hit, or a specific message pattern has been detected.

Overview

A termination condition in AgentOpera is a stateful callable object that:

1. Takes a sequence of messages as input

2. Determines whether the conversation should be terminated

Built-in Termination Conditions

AgentOpera provides several built-in termination conditions to cover common scenarios:

MaxMessageTermination

Terminates the conversation after a maximum number of messages have been exchanged:

from agentopera.chatflow.conditions import MaxMessageTermination

# Terminate after 10 messages
termination = MaxMessageTermination(max_messages=10, include_agent_event=False)

The include_agent_event parameter specifies whether AgentEvent objects are included in the message count.

TextMentionTermination

Terminates the conversation if specific text is mentioned in a message:

from agentopera.chatflow.conditions import TextMentionTermination

# Terminate when "DONE" is mentioned in any message
termination = TextMentionTermination(text="DONE")

# Terminate when "STOP" is mentioned, but only from specific agents
termination = TextMentionTermination(text="STOP", sources=["agent1", "agent2"])

TimeoutTermination

Terminates the conversation after a specified duration:

from agentopera.chatflow.conditions import TimeoutTermination

# Terminate after 60 seconds
termination = TimeoutTermination(timeout_seconds=60)

TextMessageTermination

Terminates the conversation when a TextMessage is received:

from agentopera.chatflow.conditions import TextMessageTermination

# Terminate on any TextMessage
termination = TextMessageTermination()

# Terminate on TextMessage from a specific source
termination = TextMessageTermination(source="user")

Combining Termination Conditions

AgentOpera’s termination system allows you to combine conditions using logical operators:

OR Combination

Terminate when ANY of the conditions are met:

from agentopera.chatflow.conditions import MaxMessageTermination, TextMentionTermination

# Terminate after 20 messages OR when "DONE" is mentioned
termination = MaxMessageTermination(20) | TextMentionTermination("DONE")

AND Combination

Terminate when ALL of the conditions are met:

from agentopera.chatflow.conditions import MaxMessageTermination, TextMentionTermination

# Terminate only if 10 messages have been sent AND ‘READY’ has been mentioned.
termination = MaxMessageTermination(10) & TextMentionTermination("READY")

Complete Example

Here's a complete example demonstrating how to use termination conditions with an agent team:

import asyncio
from agentopera.chatflow.conditions import MaxMessageTermination, TextMentionTermination
from agentopera.chatflow.models import OpenAIChat
from agentopera.chatflow.agents import ChatAgent
from agentopera.chatflow.teams import RoundRobinTeam

async def main():
    # Create LLM model
    llm = OpenAIChat(model="gpt-3.5-turbo")
    
    # Create agents
    researcher = ChatAgent(
        name="researcher",
        llm=llm,
        system_message="You are a research assistant who finds information on topics."
    )
    
    writer = ChatAgent(
        name="writer",
        llm=llm,
        system_message="You are a writer who crafts clear explanations based on research."
    )
    
    # Combine termination conditions
    termination = MaxMessageTermination(max_messages=10) | TextMentionTermination("TASK COMPLETE")
    
    # Create team
    team = RoundRobinTeam(
        agents=[researcher, writer],
        termination_condition=termination
    )
    
    # Run conversation
    result = await team.run(task="Research and explain quantum computing in simple terms.")
    
    # Print results
    for message in result.messages:
        print(f"{message.source}: {message.content}")
    print(f"Termination reason: {result.stop_reason}")

if __name__ == "__main__":
    asyncio.run(main())