mem0ai/mem0 · error · ValueError

When embeddings are enabled, all payloads must contain a 'da

Error message

When embeddings are enabled, all payloads must contain a 'data' field.

What it means

Raised in UpstashVector.insert when enable_embeddings=True but any payload lacks a 'data' key (or has data=None). In embedding mode the store sends the payload's 'data' text to Upstash to embed server-side instead of using the client-supplied vectors, so every record must carry the source text. The check is all-or-nothing: one malformed payload fails the whole batch.

Source

Thrown at mem0/vector_stores/upstash_vector.py:88

    def insert(
        self,
        vectors: List[list],
        payloads: Optional[List[Dict]] = None,
        ids: Optional[List[str]] = None,
    ):
        """
        Insert vectors

        Args:
            vectors (list): List of vectors to insert.
            payloads (list, optional): List of payloads corresponding to vectors. These will be passed as metadatas to the Upstash Vector client. Defaults to None.
            ids (list, optional): List of IDs corresponding to vectors. Defaults to None.
        """
        logger.info(f"Inserting {len(vectors)} vectors into namespace {self.collection_name}")

        if self.enable_embeddings:
            if not payloads or any("data" not in m or m["data"] is None for m in payloads):
                raise ValueError("When embeddings are enabled, all payloads must contain a 'data' field.")
            processed_vectors = [
                {
                    "id": ids[i] if ids else None,
                    "data": payloads[i]["data"],
                    "metadata": payloads[i],
                }
                for i, v in enumerate(vectors)
            ]
        else:
            processed_vectors = [
                {
                    "id": ids[i] if ids else None,
                    "vector": vectors[i],
                    "metadata": payloads[i] if payloads else None,
                }
                for i, v in enumerate(vectors)
            ]

View on GitHub (pinned to 001c235229)

Solutions

Ensure every metadata dict passed to add/insert includes a non-null 'data' field with the text to embed.
If your payloads use another key, remap before insert: `{**p, "data": p["text"]}`.
If you actually want client-side embeddings (mem0 computes vectors), remove enable_embeddings so the vectors argument is used instead.

Example fix

# before
store.insert(vectors=vecs, payloads=[{"memory": "prefers tea"}], ids=[...])
# enable_embeddings=True -> ValueError

# after
store.insert(vectors=vecs, payloads=[{"data": "prefers tea", "memory": "prefers tea"}], ids=[...])

Defensive patterns

Strategy: validation

Validate before calling

def validate_payloads_for_embedding(payloads):
    missing = [i for i, p in enumerate(payloads or []) if not p or p.get("data") is None]
    if payloads is None or missing:
        raise ValueError(f"payloads missing 'data' at indices {missing}; required when enable_embeddings=True")
    return payloads

Type guard

def payloads_have_data(payloads) -> bool:
    return bool(payloads) and all(
        isinstance(p, dict) and p.get("data") is not None for p in payloads
    )

Try / catch

try:
    store.insert(vectors=vectors, payloads=payloads, ids=ids)
except ValueError as e:
    if "'data' field" in str(e):
        raise ValueError("Embedding mode requires a non-null 'data' text in every payload") from e
    raise

Prevention

Standardize on a 'data' key for the embeddable text when enable_embeddings is on.
Validate the whole batch before insert — one bad payload fails all of it.
If you embed client-side, keep enable_embeddings off and pass vectors.

When it happens

Trigger: Config with `"enable_embeddings": true` (or the constructor flag) plus an `add()`/`insert()` where metadata payloads contain keys like 'memory' or 'text' but not 'data'; payloads=None; records where 'data' was set to None by upstream cleanup.

Common situations: Flipping enable_embeddings on an existing integration whose payloads predate the 'data' convention; custom embedder pipelines writing 'content' instead of 'data'; one record in a batch differing in schema from the rest.

Related errors

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/86fd6eb28f5baee9. Report an issue: GitHub.