langgenius/dify · error · ProviderNotInitializeError

provider_not_initialize

provider_not_initialize

Error message

No Embedding Model available. Please configure a valid provider in the Settings -> Model Provider.

What it means

Thrown as ProviderNotInitializeError (HTTP 400, code provider_not_initialize) when the indexing-cost estimate step cannot run because the tenant has no usable embedding model. The controller catches LLMBadRequestError from IndexingRunner.indexing_estimate and re-wraps it, masking the original detail with a generic message pointing to Settings -> Model Provider. This means the estimate never even reached the embedding call — the model resolution itself failed.

Source

Thrown at api/controllers/console/datasets/datasets.py:972

                        document_model=args["doc_form"],
                    )
                    extract_settings.append(extract_setting)
            case _:
                raise ValueError("Data source type not support")
        indexing_runner = IndexingRunner()
        try:
            response = indexing_runner.indexing_estimate(
                tenant_id=current_tenant_id,
                extract_settings=extract_settings,
                tmp_processing_rule=args["process_rule"],
                doc_form=args["doc_form"],
                doc_language=args["doc_language"],
                dataset_id=args["dataset_id"],
                indexing_technique=args["indexing_technique"],
                session=session,
            )
        except LLMBadRequestError:
            raise ProviderNotInitializeError(
                "No Embedding Model available. Please configure a valid provider in the Settings -> Model Provider."
            )
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except Exception as e:
            raise IndexingEstimateError(str(e))

        return (
            IndexingEstimateResponse(
                tokens=0,
                total_price=0,
                currency="USD",
                total_segments=response.total_segments,
                preview=response.preview,
                qa_preview=response.qa_preview,
            ).model_dump(mode="json", exclude_none=True),
            200,
        )

View on GitHub (pinned to ef8544b173)

Solutions

1. In Settings -> Model Provider, add a valid embedding model (e.g. OpenAI text-embedding-3-small) and confirm it shows as 'Available'.
2. If a provider already exists, click it and re-enter / refresh its API key, then retry the estimate.
3. Verify the embedding provider backend is reachable (curl the provider's endpoint) when using self-hosted/volc/local providers.
4. Check docker/.env / docker/envs/*.env for the correct EMBEDDING_* / provider credentials when running self-hosted Dify.

Example fix

// before: estimate called with no embedding model configured
POST /console/api/datasets/<id>/indexing-estimate  →  400 provider_not_initialize

// after: configure provider first, then estimate returns 200
// 1. UI: Settings → Model Provider → add embedding model
// 2. Retry the same estimate request

from core.model_manager import ModelManager
from models.dataset import Dataset

def has_embedding_model(tenant_id: str, dataset: Dataset) -> bool:
    """Return True only if the dataset's embedding provider is resolvable."""
    if not dataset.embedding_model_provider:
        return False
    try:
        mm = ModelManager.for_tenant(tenant_id=tenant_id)
        mm.get_model_instance(
            tenant_id=tenant_id,
            provider=dataset.embedding_model_provider,
            model_type=ModelType.TEXT_EMBEDDING,
            model=dataset.embedding_model or "",
        )
        return True
    except Exception:
        return False

# before calling indexing_estimate
if not has_embedding_model(current_tenant_id, dataset):
    return error("Configure an embedding model first.")

# controller-level: narrow the catch instead of relying on LLMBadRequestError
try:
    response = indexing_runner.indexing_estimate(...)
except (LLMBadRequestError, ProviderTokenNotInitError) as ex:
    # surface a single, actionable error to the client
    raise ProviderNotInitializeError(
        "Embedding model unavailable for this tenant. Configure one in Settings -> Model Provider."
    )

When it happens

Trigger: Calling the dataset indexing-estimate endpoint (POST /console/api/datasets/<id>/indexing-estimate or the estimate sub-route used during document creation preview) on a tenant that has either (a) no embedding model configured at all, or (b) a configured provider whose model is unavailable/invalid so model resolution raises LLMBadRequestError. Reproduced whenever the Knowledge page opens the cost-estimate panel before any embedding provider is set up.

Common situations: Fresh tenant that has never visited Settings -> Model Provider; an embedding provider was configured but its API key was later revoked or deleted; a self-hosted model server (e.g. Xinference/Ollama) went offline after the dataset was created; the dataset references a model from a provider plugin that was uninstalled.

Related errors

• provider_not_initialize
• model_currently_not_support
• server_4xx_other
• provider_quota_exceeded
• provider_not_support_speech_to_text

AI-assisted analysis of langgenius/dify@ef8544b173 (2026-08-12). Data as JSON: /api/errors/aaa1317cab0d9ef4. Report an issue: GitHub.