BerriAI/litellm · error · ValueError

Missing Azure AI API Base - Set AZURE_AI_API_BASE environmen

Error message

Missing Azure AI API Base - Set AZURE_AI_API_BASE environment variable or pass api_base parameter

What it means

Thrown by the Azure AI OCR config's validate_environment when building request headers: no API base URL could be resolved. LiteLLM needs the Azure AI services endpoint (e.g. https://<resource>.services.ai.azure.com) to know where to send the OCR request, and it looked both in the explicit api_base parameter and the AZURE_AI_API_BASE environment variable before giving up.

Source

Thrown at litellm/llms/azure_ai/ocr/transformation.py:66

        Validate environment and return headers for Azure AI OCR.

        Azure AI uses Bearer token authentication with AZURE_AI_API_KEY.
        """
        # Get API key from environment if not provided
        if api_key is None:
            api_key = get_secret_str(AZURE_AI_OCR_API_KEY_ENV_VAR)

        if api_key is None:
            raise ValueError(
                "Missing Azure AI API Key - A call is being made to Azure AI but no key is set either in the environment variables or via params"
            )

        # Validate API base is provided
        if api_base is None:
            api_base = get_secret_str("AZURE_AI_API_BASE")

        if api_base is None:
            raise ValueError(
                "Missing Azure AI API Base - Set AZURE_AI_API_BASE environment variable or pass api_base parameter"
            )

        headers = {
            "Authorization": f"Bearer {api_key}",
            "Content-Type": "application/json",
            **headers,
        }

        return headers

    def get_complete_url(
        self,
        api_base: str | None,
        model: str,
        optional_params: dict,
        litellm_params: dict | None = None,
        **kwargs,

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Set the environment variable: export AZURE_AI_API_BASE=https://<your-resource>.services.ai.azure.com
2. Or pass api_base explicitly: litellm.ocr(model='azure_ai/mistral-ocr', document=..., api_base='https://<resource>.services.ai.azure.com', api_key=...)
3. If using LiteLLM proxy, add api_base (and api_key) to the model's litellm_params in config.yaml and restart

Example fix

# before
litellm.ocr(model='azure_ai/mistral-ocr', document={'type':'document_url','document_url':url}, api_key=os.environ['AZURE_AI_API_KEY'])

# after
litellm.ocr(model='azure_ai/mistral-ocr', document={'type':'document_url','document_url':url}, api_key=os.environ['AZURE_AI_API_KEY'], api_base='https://my-resource.services.ai.azure.com')

import os
api_base = os.environ.get('AZURE_AI_API_BASE')
if not api_base:
    raise SystemExit('Configure AZURE_AI_API_BASE before running OCR jobs')

try:
    litellm.ocr(model='azure_ai/mistral-ocr', document=doc)
except ValueError as e:
    if 'Missing Azure AI API Base' in str(e):
        raise ConfigurationError('Azure AI endpoint not set') from e
    raise

When it happens

Trigger: Calling litellm.ocr(...) with an azure_ai/mistral-ocr model while neither api_base is passed (or router model_info entry) nor AZURE_AI_API_BASE is set in the environment. The header builder resolves the API key first, so a missing key fails earlier; this fires only when the key exists but the base URL does not.

Common situations: Copying an OCR example that hardcodes only the API key; deploying to a new environment (container/CI) where AZURE_AI_API_BASE was not exported; using the proxy where the model entry in config.yaml lacks api_base.

Related errors

• Azure AI API Base is required. api_base=None. Set in call or
• AZURE_SENTINEL_DCR_IMMUTABLE_ID is required. Set it as an en
• AZURE_SENTINEL_ENDPOINT is required. Set it as an environmen
• AZURE_SENTINEL_TENANT_ID or AZURE_TENANT_ID is required. Set
• AZURE_SENTINEL_CLIENT_ID or AZURE_CLIENT_ID is required. Set

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/8459c79d173e7fb7. Report an issue: GitHub.