BerriAI/litellm · error · DatabricksException

Missing API Key - A call is being made to LLM Provider but n

Error message

Missing API Key - A call is being made to LLM Provider but no key is set either in the environment variables ({LLM_PROVIDER}_API_KEY) or via params

What it means

Raised when a Databricks call is configured as a custom endpoint (custom_endpoint=True), no OAuth client credentials are present, no api_key is given, and no headers were supplied — so no authentication material exists at all. LiteLLM refuses the call with a 400 rather than sending an unauthenticated request.

Source

Thrown at litellm/llms/databricks/common_utils.py:351

        # Check for OAuth M2M credentials (recommended for production)
        client_id: Final = os.getenv("DATABRICKS_CLIENT_ID")
        client_secret: Final = os.getenv("DATABRICKS_CLIENT_SECRET")

        # Determine api_base first
        if api_base is None:
            api_base = os.getenv("DATABRICKS_API_BASE")

        if client_id and client_secret and api_base:
            # Use OAuth M2M flow (preferred for production)
            verbose_logger.debug("Using OAuth M2M authentication for Databricks")
            access_token: Final = self._get_oauth_m2m_token(api_base, client_id, client_secret)
            headers = headers or {}
            headers["Authorization"] = f"Bearer {access_token}"
            headers["Content-Type"] = "application/json"
        elif api_key is None and not headers:
            if custom_endpoint is True:
                raise DatabricksException(
                    status_code=400,
                    message="Missing API Key - A call is being made to LLM Provider but no key is set either in the environment variables ({LLM_PROVIDER}_API_KEY) or via params",
                )
            else:
                # Fallback to Databricks SDK (registers partner telemetry)
                verbose_logger.debug("Using Databricks SDK for authentication")
                api_base, headers = self._get_databricks_credentials(
                    api_base=api_base, api_key=api_key, headers=headers
                )

        if api_base is None:
            if custom_endpoint:
                raise DatabricksException(
                    status_code=400,
                    message="Missing API Base - A call is being made to LLM Provider but no api base is set either in the environment variables ({LLM_PROVIDER}_API_KEY) or via params",
                )
            else:
                api_base, headers = self._get_databricks_credentials(

View on GitHub (pinned to 6c2dcb801b)

Solutions

Set DATABRICKS_API_KEY (or the provider-prefixed variant) in the environment
Or pass api_key directly in the call / litellm_params
If the gateway uses a custom header, pass explicit headers with the Authorization entry
Verify the env var name matches the provider prefix LiteLLM expects

Example fix

# before
response = litellm.completion(model="databricks/custom-model", messages=msgs, api_base=GW_URL)

# after
response = litellm.completion(model="databricks/custom-model", messages=msgs, api_base=GW_URL, api_key=os.environ["DATABRICKS_API_KEY"])

Defensive patterns

Strategy: validation

Validate before calling

if custom_endpoint and not (api_key or headers or os.getenv("DATABRICKS_API_KEY")):
    raise RuntimeError("Custom Databricks endpoint requires an API key or explicit auth headers")

Try / catch

try:
    resp = litellm.completion(model=m, messages=msgs, api_base=gw_url)
except Exception as e:
    if "Missing API Key" in str(e):
        raise ConfigError("Provide the gateway/API key for the custom endpoint") from e
    raise

Prevention

When routing Databricks models through a gateway, store the gateway key in litellm_params, not ambient env
Add a preflight check that every custom-endpoint model declares an auth source

When it happens

Trigger: Using a custom/ai-gateway Databricks endpoint (custom endpoint flag set) while DATABRICKS_API_KEY and api_key are both absent and no Authorization header is passed manually.

Common situations: Pointing Databricks models at an internal AI gateway that requires its own key, but the key env var was never wired in; migrating config where api_base was updated but api_key was dropped.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/ecd841f4b9463ca5. Report an issue: GitHub.