BerriAI/litellm · error · HerokuError

No api base was set. Please provide an api_base, or set the

Error message

No api base was set. Please provide an api_base, or set the HEROKU_API_BASE environment variable.

What it means

Raised by HerokuChatConfig.get_complete_url when neither an explicit api_base nor the HEROKU_API_BASE environment variable is present. Unlike cloud providers with fixed endpoints, Heroku AI deployments are per-app, so LiteLLM cannot construct a URL without being told where your inference endpoint lives; it raises HerokuError instead of guessing.

Source

Thrown at litellm/llms/heroku/chat/transformation.py:70

    ) -> tuple[str | None, str | None]:
        api_base = api_base or os.getenv("HEROKU_API_BASE")
        api_key = api_key or os.getenv("HEROKU_API_KEY")

        return api_base, api_key

    def get_complete_url(
        self,
        api_base: str | None,
        api_key: str | None,
        model: str,
        optional_params: dict,
        litellm_params: dict,
        stream: bool | None = None,
    ) -> str:
        api_base, _ = self._get_openai_compatible_provider_info(api_base, api_key)

        if not api_base:
            raise HerokuError(
                "No api base was set. Please provide an api_base, or set the HEROKU_API_BASE environment variable."
            )

        if not api_base.endswith("/v1/chat/completions"):
            api_base = f"{api_base}/v1/chat/completions"

        return api_base

View on GitHub (pinned to 6c2dcb801b)

Solutions

Set the env var: export HEROKU_API_BASE=https://<your-heroku-app>.herokuapp.com
Or pass api_base per call: litellm.completion(model='heroku/...', api_base='https://<app>.herokuapp.com', api_key=...).
Confirm the URL is your Heroku inference addon endpoint (the code will append /v1/chat/completions automatically if missing), so supply the bare app URL.

Example fix

# before
litellm.completion(model='heroku/claire-3.5-sonnet', messages=msgs, api_key=os.environ['HEROKU_API_KEY'])
# raises HerokuError: No api base was set

# after
litellm.completion(
    model='heroku/claire-3.5-sonnet',
    messages=msgs,
    api_base=os.environ['HEROKU_API_BASE'],  # https://us.inference.heroku.com
    api_key=os.environ['HEROKU_API_KEY'],
)

Defensive patterns

Strategy: validation

Validate before calling

import os

def heroku_base(api_base: str | None = None) -> str:
    base = api_base or os.environ.get("HEROKU_API_BASE")
    if not base:
        raise ValueError("HEROKU_API_BASE not set — configure it before calling heroku models")
    return base

Try / catch

try:
    litellm.completion(model="heroku/...", messages=msgs, api_base=base, api_key=key)
except Exception as e:
    if "No api base was set" in str(e):
        raise RuntimeError("Missing HEROKU_API_BASE — check deployment env") from e
    raise

Prevention

Add HEROKU_API_BASE alongside HEROKU_API_KEY in every environment that calls heroku models.
Fail fast at startup: assert required env vars for each provider you enable.

When it happens

Trigger: Calling litellm.completion(model='heroku/<model>') without api_base and without HEROKU_API_BASE set. Note the code first calls _get_openai_compatible_provider_info, which reads env vars like HEROKU_API_BASE; if none resolve, api_base is falsy and this raises.

Common situations: Assuming Heroku has a shared API host like OpenAI; forgetting the endpoint URL differs per Heroku app; deploying to a new environment where HEROKU_API_BASE was not injected; passing api_key but not api_base.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/1cb7276460b791a4. Report an issue: GitHub.