Implementation:Microsoft Autogen TerminationCondition Combinators

Knowledge Sources	Microsoft AutoGen
Domains	Multi-Agent Systems, Conversation Control, Boolean Algebra, AI Agents
Last Updated	2026-02-11 00:00 GMT

Overview

Concrete tools for defining and composing conversation stopping rules provided by Microsoft AutoGen.

Description

AutoGen provides a family of termination condition classes that can be instantiated individually or composed using Python's | (OR) and & (AND) operators. The three most commonly used atomic conditions are:

TextMentionTermination: Scans each message's text content for a specific substring. When found, the conversation stops. Optionally filters by source agent name. This is the standard mechanism for agents to signal task completion (e.g., by saying "TERMINATE").

MaxMessageTermination: Counts messages across turns and stops when the total reaches a threshold. By default, only BaseChatMessage instances are counted (not agent events), but this can be changed with the include_agent_event flag.

TokenUsageTermination: Tracks cumulative token usage from model responses and stops when any of the configured limits (total, prompt, or completion tokens) is exceeded.

The | operator creates an OrTerminationCondition that stops when any constituent condition triggers. The & operator creates an AndTerminationCondition that stops only when all constituent conditions have triggered. These combinators can be chained to express arbitrarily complex stopping logic.

Usage

Import termination conditions from autogen_agentchat.conditions and pass them to team constructors via the termination_condition parameter. Compose multiple conditions with | and & operators before passing.

Code Reference

Source Location

Repository: Microsoft AutoGen
File (atomic conditions): python/packages/autogen-agentchat/src/autogen_agentchat/conditions/_terminations.py
File (combinators): python/packages/autogen-agentchat/src/autogen_agentchat/base/_termination.py (lines 79-85)

Signature

class TextMentionTermination(TerminationCondition):
    def __init__(self, text: str, sources: Sequence[str] | None = None) -> None:
        ...

class MaxMessageTermination(TerminationCondition):
    def __init__(self, max_messages: int, include_agent_event: bool = False) -> None:
        ...

class TokenUsageTermination(TerminationCondition):
    def __init__(
        self,
        max_total_token: int | None = None,
        max_prompt_token: int | None = None,
        max_completion_token: int | None = None,
    ) -> None:
        ...

# Combinator operators (defined on TerminationCondition base class):
# condition_a | condition_b  -> OrTerminationCondition
# condition_a & condition_b  -> AndTerminationCondition

Import

from autogen_agentchat.conditions import TextMentionTermination, MaxMessageTermination, TokenUsageTermination

I/O Contract

Inputs

TextMentionTermination:

Name	Type	Required	Description
text	str	Yes	The substring to search for in message content. When found, the conversation stops.
sources	Sequence[str] or None	No	If provided, only messages from these agent names are checked. If None, all messages are checked.

MaxMessageTermination:

Name	Type	Required	Description
max_messages	int	Yes	Maximum number of messages allowed before stopping.
include_agent_event	bool	No	Whether to count BaseAgentEvent messages in addition to BaseChatMessage. Defaults to False.

TokenUsageTermination:

Name	Type	Required	Description
max_total_token	int or None	No	Maximum cumulative total tokens (prompt + completion). At least one of the three limits must be provided.
max_prompt_token	int or None	No	Maximum cumulative prompt tokens.
max_completion_token	int or None	No	Maximum cumulative completion tokens.

Outputs

Name	Type	Description
instance	TerminationCondition	A termination condition instance (or composed condition) to pass to a team's termination_condition parameter.
__call__ result	StopMessage or None	When evaluated, returns a StopMessage with a descriptive reason if the condition is met, or None if the conversation should continue.

Usage Examples

Basic Example

from autogen_agentchat.conditions import TextMentionTermination, MaxMessageTermination

# Stop when the agent says "TERMINATE"
text_termination = TextMentionTermination("TERMINATE")

# Stop after 10 messages
max_msg_termination = MaxMessageTermination(max_messages=10)

# Combine with OR: stop when EITHER condition is met
termination = text_termination | max_msg_termination

Token Budget with Text Signal

from autogen_agentchat.conditions import (
    TextMentionTermination,
    MaxMessageTermination,
    TokenUsageTermination,
)

# Stop on "DONE" from the summarizer agent, OR after 50k total tokens, OR after 30 messages
termination = (
    TextMentionTermination("DONE", sources=["summarizer"])
    | TokenUsageTermination(max_total_token=50000)
    | MaxMessageTermination(max_messages=30)
)

AND Composition

from autogen_agentchat.conditions import TextMentionTermination

# Stop only when BOTH the reviewer AND the editor have said "APPROVED"
termination = (
    TextMentionTermination("APPROVED", sources=["reviewer"])
    & TextMentionTermination("APPROVED", sources=["editor"])
)

Related Pages

Implements Principle

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment