langchain-ai/langchain · error · NotImplementedError

_HashedDocument is an internal abstraction that was deprecat

Error message

_HashedDocument is an internal abstraction that was deprecated in  langchain-core 0.3.63. This abstraction is marked as private and  should not have been used directly. If you are seeing this error, please  update your code appropriately.

What it means

`_HashedDocument` was a private helper in the indexing API that some users imported anyway. In langchain-core 0.3.63 its logic moved into `_get_document_with_hash` and the class was turned into a stub whose `__init__` always raises `NotImplementedError`. The class is kept importable purely so old imports do not break with an ImportError, but instantiating it is now a hard failure.

Source

Thrown at libs/core/langchain_core/indexing/api.py:244

        # Assign a unique identifier based on the hash.
        id=hash_,
        page_content=document.page_content,
        metadata=document.metadata,
    )


# This internal abstraction was imported by the langchain package internally, so
# we keep it here for backwards compatibility.
class _HashedDocument:
    def __init__(self, *args: Any, **kwargs: Any) -> None:
        """Raise an error if this class is instantiated."""
        msg = (
            "_HashedDocument is an internal abstraction that was deprecated in "
            " langchain-core 0.3.63. This abstraction is marked as private and "
            " should not have been used directly. If you are seeing this error, please "
            " update your code appropriately."
        )
        raise NotImplementedError(msg)


def _delete(
    vector_store: VectorStore | DocumentIndex,
    ids: list[str],
) -> None:
    """Delete documents from a vector store or document index by their IDs.

    Args:
        vector_store: The vector store or document index to delete from.
        ids: List of document IDs to delete.

    Raises:
        IndexingException: If the delete operation fails.
        TypeError: If the `vector_store` is neither a `VectorStore` nor a
            `DocumentIndex`.
    """
    if isinstance(vector_store, VectorStore):

View on GitHub (pinned to e32fa9a52e)

Solutions

1. Replace direct construction with a call to the public indexing API — `index()` hashes documents internally.
2. If you need a hashed document yourself, compute it via a callable passed as `key_encoder`, or hash content+metadata with hashlib in your own code.
3. Pin langchain-core < 0.3.63 only as a last-resort temporary stopgap while migrating.

Example fix

# before
hashed = _HashedDocument(page_content=doc.page_content)

# after
# let the indexer hash; or roll your own
import hashlib, uuid
uid = uuid.uuid5(uuid.NAMESPACE_URL, doc.page_content)
hashed = doc.model_copy(update={"id": str(uid)})

import langchain_core.indexing.api as api
if hasattr(api, "_get_document_with_hash"):
    ...  # new API available; do not touch _HashedDocument

try:
    hashed = _HashedDocument(...)  # legacy path
except NotImplementedError:
    hashed = None  # fall back to letting index() hash internally

When it happens

Trigger: `from langchain_core.indexing.api import _HashedDocument` followed by `_HashedDocument(...)`; older code or third-party packages (including some langchain-classic internals) that constructed it directly to pre-hash documents.

Common situations: Upgrading langchain-core past 0.3.63 with code that pre-hashed documents before calling index(); vendored tutorials or notebooks referencing the private class; CopyPasta from the indexing module's internals.

Related errors

• A pending deprecation cannot have a scheduled removal
• Cannot specify both alternative and alternative_import
• alternative_import must be a fully qualified module path. Go
• Field {obj} must have a name to be deprecated.
• {f.__name__}() got multiple values for argument {new!r}

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/2e062ebcfad0ed0a. Report an issue: GitHub.