microsoft/semantic-kernel · error · ServiceInitializationError

The NVIDIA API key is required.

Error message

The NVIDIA API key is required.

What it means

Raised as ServiceInitializationError when neither a pre-built `client` nor a usable api_key was supplied: `if not client and not nvidia_settings.api_key`. NVIDIA requires authentication for its hosted endpoint, and the connector refuses to build a client with no credentials. Note NvidiaSettings may allow a None api_key (it is optional), so this guard is the real enforcement.

Source

Thrown at python/semantic_kernel/connectors/ai/nvidia/services/nvidia_chat_completion.py:99

            env_file_path (str | None): Use the environment settings file as a fallback
                to environment variables. (Optional)
            env_file_encoding (str | None): The encoding of the environment settings file. (Optional)
            instruction_role (Literal["system", "user", "assistant", "developer"] | None): The role to use for
                'instruction' messages. Defaults to "system". (Optional)
        """
        try:
            nvidia_settings = NvidiaSettings(
                api_key=api_key,
                base_url=base_url,
                chat_model_id=ai_model_id,
                env_file_path=env_file_path,
                env_file_encoding=env_file_encoding,
            )
        except ValidationError as ex:
            raise ServiceInitializationError("Failed to create NVIDIA settings.", ex) from ex

        if not client and not nvidia_settings.api_key:
            raise ServiceInitializationError("The NVIDIA API key is required.")
        if not nvidia_settings.chat_model_id:
            # Default fallback model
            nvidia_settings.chat_model_id = DEFAULT_NVIDIA_CHAT_MODEL
            logger.warning(f"Default chat model set as: {nvidia_settings.chat_model_id}")

        # Create client if not provided
        if not client:
            client = AsyncOpenAI(
                api_key=nvidia_settings.api_key.get_secret_value() if nvidia_settings.api_key else None,
                base_url=nvidia_settings.base_url,
            )

        super().__init__(
            ai_model_id=nvidia_settings.chat_model_id,
            api_key=nvidia_settings.api_key.get_secret_value() if nvidia_settings.api_key else None,
            base_url=nvidia_settings.base_url,
            service_id=service_id or "",
            ai_model_type=NvidiaModelTypes.CHAT,

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Pass api_key explicitly: NvidiaChatCompletion(api_key=os.environ['NVIDIA_API_KEY']).
2. Set NVIDIA_API_KEY in env or .env at the project root.
3. If using a custom endpoint, pass client=AsyncOpenAI(api_key=..., base_url=...) instead of relying on env.
4. Double-check the exact env var name spelling against NvidiaSettings.

Example fix

# before
svc = NvidiaChatCompletion(ai_model_id='meta/llama3-8b-instruct')

# after
svc = NvidiaChatCompletion(
    ai_model_id='meta/llama3-8b-instruct',
    api_key=os.environ['NVIDIA_API_KEY'],
)

api_key = os.environ.get('NVIDIA_API_KEY')
assert api_key, 'NVIDIA_API_KEY is required for NvidiaChatCompletion'
svc = NvidiaChatCompletion(api_key=api_key, ai_model_id='meta/llama3-8b-instruct')

def has_nvidia_credentials(client=None, **kw) -> bool:
    return bool(client or kw.get('api_key') or os.environ.get('NVIDIA_API_KEY'))

from semantic_kernel.exceptions import ServiceInitializationError
try:
    svc = NvidiaChatCompletion(ai_model_id='meta/llama3-8b-instruct')
except ServiceInitializationError as e:
    if 'API key is required' in str(e):
        svc = NvidiaChatCompletion(api_key=os.environ['NVIDIA_API_KEY'], ai_model_id='meta/llama3-8b-instruct')
    else:
        raise

When it happens

Trigger: Constructing `NvidiaChatCompletion(...)` with no `client=` argument AND no NVIDIA_API_KEY in env/.env AND no `api_key=` arg. Also when api_key is present but empty/fails SecretStr and is dropped to None.

Common situations: Forgetting to export NVIDIA_API_KEY in the shell running the app, .env not loaded, key set in a different env name (e.g. NGC_API_KEY or NVAPI_KEY), CI runner missing the secret.

Related errors

• Failed to create NVIDIA settings.
• {type(self)} service failed to complete the chat
• Failed to create NVIDIA settings.
• Failed to create Ollama settings.
• Ollama chat model ID is required.

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/ed202192c2d4b071. Report an issue: GitHub.