microsoft/autogen · error · ValueError

At least one of max_total_token, max_prompt_token, or max_co

Error message

At least one of max_total_token, max_prompt_token, or max_completion_token must be provided

What it means

TokenUsageTermination's constructor requires at least one token budget: max_total_token, max_prompt_token, or max_completion_token. Constructing it with no arguments (all None) raises ValueError immediately — it is a usage error, not a runtime state error.

Source

Thrown at python/packages/autogen-agentchat/src/autogen_agentchat/conditions/_terminations.py:257

        max_total_token: The maximum total number of tokens allowed in the conversation.
        max_prompt_token: The maximum number of prompt tokens allowed in the conversation.
        max_completion_token: The maximum number of completion tokens allowed in the conversation.

    Raises:
        ValueError: If none of max_total_token, max_prompt_token, or max_completion_token is provided.
    """

    component_config_schema = TokenUsageTerminationConfig
    component_provider_override = "autogen_agentchat.conditions.TokenUsageTermination"

    def __init__(
        self,
        max_total_token: int | None = None,
        max_prompt_token: int | None = None,
        max_completion_token: int | None = None,
    ) -> None:
        if max_total_token is None and max_prompt_token is None and max_completion_token is None:
            raise ValueError(
                "At least one of max_total_token, max_prompt_token, or max_completion_token must be provided"
            )
        self._max_total_token = max_total_token
        self._max_prompt_token = max_prompt_token
        self._max_completion_token = max_completion_token
        self._total_token_count = 0
        self._prompt_token_count = 0
        self._completion_token_count = 0

    @property
    def terminated(self) -> bool:
        return (
            (self._max_total_token is not None and self._total_token_count >= self._max_total_token)
            or (self._max_prompt_token is not None and self._prompt_token_count >= self._max_prompt_token)
            or (self._max_completion_token is not None and self._completion_token_count >= self._max_completion_token)
        )

    async def __call__(self, messages: Sequence[BaseAgentEvent | BaseChatMessage]) -> StopMessage | None:

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Pass at least one limit, e.g. TokenUsageTermination(max_total_token=3000)
2. Check parameter spelling: max_total_token / max_prompt_token / max_completion_token (singular 'token')
3. Validate config payloads before constructing: require at least one non-None budget key

Example fix

# before
cond = TokenUsageTermination()  # ValueError

# after
cond = TokenUsageTermination(max_total_token=3000)

budget = {"max_total_token": cfg.get("max_total_token"),
         "max_prompt_token": cfg.get("max_prompt_token"),
         "max_completion_token": cfg.get("max_completion_token")}
if all(v is None for v in budget.values()):
    raise ValueError("config must set at least one token budget")

When it happens

Trigger: TokenUsageTermination() with empty parentheses; passing budgets as keyword arguments with typos (max_total_tokens with trailing 's') so all real parameters stay None; building conditions from config where the budget fields are missing.

Common situations: Copy-pasting from examples that show a budget and then stripping it; config-driven condition construction where the limits key is absent or misnamed.

Related errors

• Unsupported config type {config.GetType()}
• Messages should not be provided in options
• The agent name must be a valid Python identifier.
• model_client.model_info must be provided when max_retries_on
• Specified model_client doesn't support structured output mod

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/4d3bbaac01b808be. Report an issue: GitHub.