mlflow/mlflow · error · ValueError

Unknown Gateway vendor: {vendor!r}

Error message

Unknown Gateway vendor: {vendor!r}

What it means

ensure_gateway_connection(vendor, api_key) looks up the vendor in _GATEWAY_VENDOR_MODELS to find the Gateway model definition for the Assistant's LLM connection. If the vendor string is not a registered key, it raises ValueError 'Unknown Gateway vendor: ...'. Only vendors with an explicit mapping are supported.

Source

Thrown at mlflow/assistant/gateway_connection.py:24

from mlflow.tracking._tracking_service.utils import _get_store

_GATEWAY_VENDOR_MODELS = {
    "openai": "gpt-5.5",
    "anthropic": "claude-sonnet-5",
    "gemini": "gemini-3-pro",
}

_NOT_FOUND = ErrorCode.Name(RESOURCE_DOES_NOT_EXIST)


class GatewayUnsupportedError(Exception):
    """Raised when the tracking store has no AI Gateway support."""


def ensure_gateway_connection(vendor: str, api_key: str) -> str:
    """Create or rotate the Gateway resources for an Assistant vendor key."""
    if (model_name := _GATEWAY_VENDOR_MODELS.get(vendor)) is None:
        raise ValueError(f"Unknown Gateway vendor: {vendor!r}")

    name = f"mlflow-assistant-{vendor}"
    store = _get_store()

    try:
        try:
            secret = store.get_secret_info(secret_name=name)
        except MlflowException as e:
            if e.error_code != _NOT_FOUND:
                raise
            secret = store.create_gateway_secret(
                secret_name=name,
                secret_value={"api_key": api_key},
                provider=vendor,
            )
        else:
            store.update_gateway_secret(
                secret_id=secret.secret_id,

View on GitHub (pinned to 6a27f2decc)

Solutions

1. Check the exact supported vendor strings in _GATEWAY_VENDOR_MODELS in mlflow/assistant/gateway_connection.py and use one verbatim.
2. Fix casing/typos/whitespace in the vendor value in your config or CLI argument.
3. Upgrade MLflow if the vendor is supported only in newer releases.

Example fix

// before
ensure_gateway_connection("Anthropic ", key)
// after
ensure_gateway_connection("anthropic", key)

from mlflow.assistant.gateway_connection import _GATEWAY_VENDOR_MODELS
assert vendor in _GATEWAY_VENDOR_MODELS, f"Unsupported vendor: {vendor!r}"

def is_supported_vendor(vendor):
    return vendor in _GATEWAY_VENDOR_MODELS

try:
    ensure_gateway_connection(vendor, api_key)
except ValueError as e:
    if "Unknown Gateway vendor" in str(e):
        raise SystemExit(f"{e}. Supported: {sorted(_GATEWAY_VENDOR_MODELS)}")
    raise

When it happens

Trigger: Calling ensure_gateway_connection (directly or via _store_gateway_api_key, e.g. `mlflow assistant` key setup) with a vendor name not in _GATEWAY_VENDOR_MODELS — typos ('anthropic ' with whitespace, 'Anthropic' casing mismatch) or a genuinely unsupported vendor.

Common situations: Config file naming a provider that this MLflow version's Assistant doesn't support; case/whitespace mistakes in vendor string; using a vendor added in a newer MLflow release than the installed one.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• project_path is required for 'project' skills location
• custom_path is required for 'custom' skills location
• This MLflow server's tracking backend does not support the A
• Provider configuration not found for {name}
• {self._display_name} is not configured. {self._connection_hi

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/47a9a413e83975e2. Report an issue: GitHub.