BerriAI/litellm · error · ValueError

Azure AI API Base is required. api_base=None. Set in call or

Error message

Azure AI API Base is required. api_base=None. Set in call or via `AZURE_AI_API_BASE` env var.

What it means

Azure AI rerank reuses the Cohere rerank spec but targets your Azure AI resource, so it needs the resource endpoint. get_complete_url raises this when api_base is None — no fallback to a default host exists. Set it per call or via AZURE_AI_API_BASE.

Source

Thrown at litellm/llms/azure_ai/rerank/transformation.py:29

from litellm.llms.cohere.rerank.transformation import CohereRerankConfig
from litellm.secret_managers.main import get_secret_str
from litellm.types.utils import RerankResponse
from litellm.utils import _add_path_to_api_base


class AzureAIRerankConfig(CohereRerankConfig):
    """
    Azure AI Rerank - Follows the same Spec as Cohere Rerank
    """

    def get_complete_url(
        self,
        api_base: str | None,
        model: str,
        optional_params: dict | None = None,
    ) -> str:
        if api_base is None:
            raise ValueError(
                "Azure AI API Base is required. api_base=None. Set in call or via `AZURE_AI_API_BASE` env var."
            )
        original_url: Final = httpx.URL(api_base)
        if not original_url.is_absolute_url:
            raise ValueError(
                "Azure AI API Base must be an absolute URL including scheme (e.g. "
                "'https://<resource>.services.ai.azure.com'). "
                f"Got api_base={api_base!r}."
            )
        normalized_path: Final = original_url.path.rstrip("/")

        # Allow callers to pass either full v1/v2 rerank endpoints:
        # - https://<resource>.services.ai.azure.com/v1/rerank
        # - https://<resource>.services.ai.azure.com/providers/cohere/v2/rerank
        if normalized_path.endswith("/v1/rerank") or normalized_path.endswith("/v2/rerank"):
            return str(original_url.copy_with(path=normalized_path or "/"))

        # If callers pass just the version path (e.g. ".../v2" or ".../providers/cohere/v2"), append "/rerank"

View on GitHub (pinned to 6c2dcb801b)

Solutions

export AZURE_AI_API_BASE=https://<resource>.services.ai.azure.com
Or pass api_base='https://<resource>.services.ai.azure.com' to litellm.rerank(...)
In the proxy, set litellm_params.api_base for the azure_ai rerank model and reload config

Example fix

# before
litellm.rerank(model='azure_ai/cohere-rerank-v3.5', query='hello', documents=['a','b'])

# after
litellm.rerank(model='azure_ai/cohere-rerank-v3.5', query='hello', documents=['a','b'], api_base='https://my-resource.services.ai.azure.com', api_key=os.environ['AZURE_AI_API_KEY'])

Defensive patterns

Strategy: validation

Validate before calling

api_base = os.environ.get('AZURE_AI_API_BASE')
assert api_base, 'AZURE_AI_API_BASE must be set for azure_ai rerank'

Prevention

Set AZURE_AI_API_BASE and AZURE_AI_API_KEY together in deployment env
Wrap azure_ai rerank calls in a thin client function that injects api_base

When it happens

Trigger: Calling litellm.rerank(model='azure_ai/<cohere-rerank-model>', query=..., documents=[...]) without api_base in the call, router model_info, or the AZURE_AI_API_BASE environment variable.

Common situations: Switching from cohere/rerank-v3.5 to azure_ai/... and forgetting the endpoint is not derivable from the model name; proxy config.yaml model entry missing api_base; local .env not loaded in the deployment.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/aafb658d5996e7e3. Report an issue: GitHub.