unslothai/unsloth · error · ValueError

Add a Provider connection block before running this recipe.

Error message

Add a Provider connection block before running this recipe.

What it means

ValueError raised by _validate_recipe_runtime_support when a recipe declares at least one LLM column (column_type starting with 'llm-') but the model_providers list is empty. LLM columns need a configured model provider connection to execute; with none, the recipe cannot run, so validation fails fast before a job is spawned.

Source

Thrown at studio/backend/core/data_recipe/service.py:166

            )
        )

    return providers


def _recipe_has_llm_columns(recipe: dict[str, Any]) -> bool:
    for column in recipe.get("columns", []):
        if not isinstance(column, dict):
            continue
        column_type = column.get("column_type")
        if isinstance(column_type, str) and column_type.startswith("llm-"):
            return True
    return False


def _validate_recipe_runtime_support(recipe: dict[str, Any], model_providers: list[Any]) -> None:
    if _recipe_has_llm_columns(recipe) and not model_providers:
        raise ValueError("Add a Provider connection block before running this recipe.")


def recipe_has_stdio_mcp(recipe: dict[str, Any]) -> bool:
    """True when the recipe asks for a local (stdio) MCP provider, i.e. a command
    this host would run. Routes gate on it to keep that behind a UI session."""
    providers = recipe.get("mcp_providers") or []
    if not isinstance(providers, list):
        return False
    return any(
        isinstance(provider, dict) and provider.get("provider_type") == "stdio"
        for provider in providers
    )


def build_mcp_providers(recipe: dict[str, Any]) -> list:
    from data_designer.config.mcp import LocalStdioMCPProvider, MCPProvider  # pyright: ignore[reportMissingImports]

    # Same gate as the chat MCP path: stdio providers spawn a local subprocess,

View on GitHub (pinned to 203007d190)

Solutions

Add a Provider connection block (API key/endpoint for the model backend) in the recipe or workspace settings, then re-run.
If calling the API programmatically, include the configured model providers in the run payload.
Or remove/replace the LLM columns if no LLM generation is intended.

Example fix

# before
recipe = {"columns": [{"column_type": "llm-generate", "prompt": "..."}]}
run_recipe(recipe, model_providers=[])

# after
recipe = {
    "columns": [{"column_type": "llm-generate", "prompt": "..."}],
    "mcp_providers": [],
}
run_recipe(recipe, model_providers=[{"name": "openai", "api_key": os.environ['OPENAI_API_KEY']}])

Defensive patterns

Strategy: validation

Validate before calling

def recipe_has_llm_columns(recipe):
    return any(
        isinstance(c, dict) and isinstance(c.get('column_type'), str) and c['column_type'].startswith('llm-')
        for c in recipe.get('columns', [])
    )
assert not (recipe_has_llm_columns(recipe) and not model_providers), 'add a Provider connection first'

Type guard

def recipe_has_llm_columns(recipe: dict) -> bool: ...  # as above; use to branch UI/validation

Try / catch

try:
    run_recipe(recipe, model_providers=providers)
except ValueError as e:
    if 'Provider connection' in str(e):
        prompt_user_to_add_provider(); return

Prevention

Validate LLM-column vs provider presence client-side before submit.
Keep provider connections configured at the workspace level so recipes inherit them.
Re-check provider connections after revoking or rotating keys.

When it happens

Trigger: Submitting a recipe with columns whose column_type is e.g. 'llm-judge' or 'llm-generate' while no Provider connection blocks exist in the runtime; providers configured at the workspace level but not passed into the run request; deleting/disconnecting all providers then re-running an old recipe.

Common situations: New users building their first recipe with an LLM column without adding a Provider connection; provider tokens revoked or connections removed after the recipe was authored; API callers constructing run payloads that omit providers.

Related errors

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/29d934de4df50b37. Report an issue: GitHub.