chroma-core/chroma · error · ValueError

You must provide an embedding function to compute embeddings

Error message

You must provide an embedding function to compute embeddings.https://docs.trychroma.com/guides/embeddings

What it means

`_embed` needs an embedding function to convert text into vectors. It first checks the collection schema's configured dense embedding function (float_list vector_index config), then the collection's `embedding_function` argument; when both are absent and text inputs are given, it raises. The linked docs (https://docs.trychroma.com/guides/embeddings) describe the supported options.

Source

Thrown at chromadb/api/models/CollectionCommon.py:792

                    override.float_list.vector_index.config.embedding_function,
                )
            elif (
                schema.defaults.float_list is not None
                and schema.defaults.float_list.vector_index is not None
                and schema.defaults.float_list.vector_index.config.embedding_function
                is not None
            ):
                schema_embedding_function = cast(
                    EmbeddingFunction[Embeddable],
                    schema.defaults.float_list.vector_index.config.embedding_function,
                )

        if schema_embedding_function is not None:
            if is_query and hasattr(schema_embedding_function, "embed_query"):
                return schema_embedding_function.embed_query(input=input)
            return schema_embedding_function(input=input)
        if self._embedding_function is None:
            raise ValueError(
                "You must provide an embedding function to compute embeddings."
                "https://docs.trychroma.com/guides/embeddings"
            )
        if is_query:
            return self._embedding_function.embed_query(input=input)
        else:
            return self._embedding_function(input=input)

    def _sparse_embed(
        self,
        input: Any,
        sparse_embedding_function: SparseEmbeddingFunction[Any],
        is_query: bool = False,
    ) -> Any:
        if is_query:
            return sparse_embedding_function.embed_query(input=input)
        return sparse_embedding_function(input=input)

View on GitHub (pinned to aecdd12c8a)

Solutions

Pass an embedding function at creation: `client.get_or_create_collection(name, embedding_function=MyEF())`
Or configure it in the collection schema's dense (float_list) vector index config
Or send precomputed `embeddings` so no client-side function is needed
If a default EF is desired, pass it explicitly (e.g. ONNXMiniLM_L6_V2) rather than relying on implicit defaults

Example fix

# before
col = client.get_or_create_collection(name="docs")
col.add(documents=["hello"])  # ValueError: no embedding function

# after
from chromadb.utils.embedding_functions import DefaultEmbeddingFunction
col = client.get_or_create_collection(name="docs", embedding_function=DefaultEmbeddingFunction())
col.add(documents=["hello"])

Defensive patterns

Strategy: validation

Validate before calling

ef_configured = collection.embedding_function is not None or schema_has_dense_ef(collection)
if sending_text and not ef_configured:
    raise ValueError("configure an embedding function or send precomputed embeddings")

Prevention

Always pass embedding_function= (or a schema dense EF) when creating collections that will embed text
After upgrading Chroma, audit collections created without an explicit EF
Centralize collection creation so the EF decision cannot be silently skipped

When it happens

Trigger: Collection created with neither `embedding_function=` nor a schema dense vector config carrying an embedding function, then add/upsert/query with textual input that requires embedding.

Common situations: Upgrading to a version where the default embedding function is no longer implicitly downloaded/attached; creating collections via a client path that does not accept or forward an embedding function.

Related errors

AI-assisted analysis of chroma-core/chroma@aecdd12c8a (2026-08-16). Data as JSON: /api/errors/2b5178ab4ba82bf4. Report an issue: GitHub.