Continuation after reaching max output tokens?

[preyneyv]

I want to be able to prompt the agent to "continue" when it reaches its token limit. This is to support a use case of long text parsing/generation, where we occasionally hit the 32k output token limit on GPT-4.1.

Here's some code that consistently hits the token limit.

import asyncio

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.messages import TextMessage
from autogen_core import CancellationToken
from autogen_core.models import ModelFamily, ModelInfo
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(
    model="gpt-4.1-mini-2025-04-14",
    model_info=ModelInfo(
        family=ModelFamily.GPT_4O,
        function_calling=True,
        json_output=True,
        vision=False,
    ),
    temperature=0,
    max_tokens=10,
)

agent = AssistantAgent(name="assistant", model_client=model_client, system_message="You are a helpful assistant.")

async def main() -> None:
    async for message in agent.on_messages_stream(
        [TextMessage(content="write a poem about airplanes", source="user")],
        cancellation_token=CancellationToken(),
    ):
        print(message)

asyncio.run(main())

By subclassing AssistantAgent, I've found that there is a finish_reason on model_result that gets set to "length" when the token limit is reached.

Ideally, if that's the case, I want to prompt the model to continue, but I don't see this finish reason propagated to the caller.

What's the right approach to this?

[ekzhu]

Use custom model context for your agent helps: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/agents.html#using-model-context