langchain-ai/langchain · error · NotImplementedError

`add_documents` and `add_texts` has not been implemented for

Error message

`add_documents` and `add_texts` has not been implemented for {self.__class__.__name__}

What it means

Raised by the base `VectorStore.add_documents` when the subclass implements neither `add_texts` nor `add_documents`/`upsert` override, meaning the store has no ingestion path at all. The message names both methods because the shim tries `add_texts` as the fallback primitive before giving up.

Source

Thrown at libs/core/langchain_core/vectorstores/base.py:263

            List of IDs of the added texts.
        """
        if type(self).add_texts != VectorStore.add_texts:
            if "ids" not in kwargs:
                ids = [doc.id for doc in documents]

                # If there's at least one valid ID, we'll assume that IDs
                # should be used.
                if any(ids):
                    kwargs["ids"] = ids

            texts = [doc.page_content for doc in documents]
            metadatas = [doc.metadata for doc in documents]
            return self.add_texts(texts, metadatas, **kwargs)
        msg = (
            f"`add_documents` and `add_texts` has not been implemented "
            f"for {self.__class__.__name__} "
        )
        raise NotImplementedError(msg)

    async def aadd_documents(
        self, documents: list[Document], **kwargs: Any
    ) -> list[str]:
        """Async run more documents through the embeddings and add to the `VectorStore`.

        Args:
            documents: Documents to add to the `VectorStore`.
            **kwargs: Additional keyword arguments.

        Returns:
            List of IDs of the added texts.
        """
        # If the async method has been overridden, we'll use that.
        if type(self).aadd_texts != VectorStore.aadd_texts:
            if "ids" not in kwargs:
                ids = [doc.id for doc in documents]

View on GitHub (pinned to e32fa9a52e)

Solutions

Implement `add_texts` (preferred minimal primitive) or `add_documents`/`upsert` in the subclass.
If the store is populated externally (e.g. by a separate indexer), remove ingestion calls from your pipeline and load documents out-of-band.
Gate generic pipelines: only call `add_documents` when `type(store).add_texts is not VectorStore.add_texts`.

Example fix

# before
class ReadOnlyStore(VectorStore):
    def similarity_search(self, query, k=4, **kwargs):
        return self._search(query, k)

ReadOnlyStore(...).add_documents(docs)  # NotImplementedError

# after
class WritableStore(ReadOnlyStore):
    def add_texts(self, texts, metadatas=None, **kwargs):
        return self._backend.bulk_insert(texts, metadatas or [{}] * len(texts))

Defensive patterns

Strategy: type-guard

Validate before calling

from langchain_core.vectorstores import VectorStore

def ingestion_ready(store: VectorStore) -> bool:
    return (
        type(store).add_texts is not VectorStore.add_texts
        or type(store).add_documents is not VectorStore.add_documents
    )

if ingestion_ready(store):
    ids = store.add_documents(docs)
else:
    raise RuntimeError(f"{type(store).__name__} is read-only; load documents externally")

Type guard

from langchain_core.vectorstores import VectorStore

def can_add_documents(store: VectorStore) -> bool:
    """True if any ingestion path is implemented."""
    return (
        type(store).add_texts is not VectorStore.add_texts
        or type(store).add_documents is not VectorStore.add_documents
    )

Try / catch

try:
    ids = store.add_documents(docs)
except NotImplementedError as e:
    raise RuntimeError(
        f"Store {type(store).__name__} cannot ingest; populate it via its native loader"
    ) from e

Prevention

Feature-detect ingestion before calling `from_documents` helpers.
For externally-populated indexes, wrap them in a read-only façade that hides ingestion APIs.
When subclassing, implementing `add_texts` alone satisfies both `add_texts` and `add_documents`.

When it happens

Trigger: Calling `add_documents(docs)` or `VectorStore.from_documents(docs, store)` on a custom subclass that only implements search/read methods; stores that subclass `VectorStore` purely for interface compliance.

Common situations: Read-only or externally-populated index wrappers; prototype subclasses where only `similarity_search` was written; helper utilities that blindly call `from_documents` on any `VectorStore` instance.

Related errors

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/0ac0204ff9d1d376. Report an issue: GitHub.