666ghj/MiroFish · error · RuntimeError

Zep batch item submission failed

Error message

Zep batch {batch_id} item submission failed

What it means

RuntimeError in submit_document_batch: client.batch.add failed with an error that is_retryable_zep_error classified as non-retryable (e.g. 400 validation, 401 auth, 404 unknown batch). Since replaying a non-idempotent add against an ambiguous state is unsafe and recovery reconciliation is only attempted for retryable failures, the code fails immediately with the original error as cause.

Solutions

1. Read the chained cause exception — it carries the true Zep error code and message.
2. 401/403: refresh ZEP_API_KEY and restart the build; the batch journal allows clean resume.
3. 404/409 on the batch: the batch no longer accepts items — clear project.zep_batch_id/operation_id and start a new batch.
4. 400/validation: fix the payload (re-check chunk sizes/counts with validate_batch_chunks) before resubmitting.
5. Pin the SDK version if a client upgrade changed the add-item schema.

Example fix

# before
else:
    raise RuntimeError(f"Zep batch {batch_id} item submission failed") from e

# after - classify the refusal with the underlying status for faster diagnosis
else:
    status = getattr(getattr(e, 'response', None), 'status_code', None)
    raise RuntimeError(
        f"Zep batch {batch_id} item submission failed"
        + (f" (HTTP {status})" if status else "")
        + "; batch may be deleted, processed, or the payload rejected"
    ) from e

builder.validate_batch_chunks(chunks, batch_size=batch_size)  # catches size/count issues pre-flight
assert graph_id and project.zep_batch_id, 'batch must exist before adding items'

try:
    submission = builder.submit_document_batch(graph_id, chunks, batch_size=bs)
except RuntimeError as e:
    cause = e.__cause__
    status = getattr(getattr(cause, 'response', None), 'status_code', None)
    if status in (401, 403):
        logger.error('Zep auth failed during item add; check ZEP_API_KEY')
    elif status in (404, 409):
        project.zep_batch_id = None  # batch gone; next run starts a new batch
    raise

When it happens

Trigger: Adding items to a batch_id that was deleted or already processed (404/409); auth token expired mid-build (401); item payload violates Zep limits (e.g. a chunk >10,000 chars slipping past validation); malformed request after SDK version change (400).

Common situations: ZEP_API_KEY rotated or revoked during a long build; the batch was manually deleted in the Zep console mid-run; SDK upgrade changing add-item request shape; chunker regression producing oversized chunks.

Related errors

• Multiple Zep batches match operation
• Persisted Zep batch does not match the current graph input
• Zep Batch API returned no batch_id
• Zep batch acknowledged of items
• Zep batch item submission is unconfirmed; the draft was not…

AI-assisted analysis of 666ghj/MiroFish@b5b53acc57 (2026-08-14). Data as JSON: /api/errors/4d424d11d246f77e. Report an issue: GitHub.

Appendix: source

Thrown at backend/app/services/graph_builder.py:517

                        batch_id,
                        expected_item_count,
                    )
                    recovered_indexes = {
                        getattr(item, "sequence_index", None)
                        for item in recovered_items
                    }
                    if (
                        len(recovered_items) == expected_item_count
                        and recovered_indexes == set(range(expected_item_count))
                    ):
                        item_details = recovered_items[i:expected_item_count]
                    else:
                        raise RuntimeError(
                            f"Zep batch {batch_id} item submission is unconfirmed; "
                            "the draft was not processed or replayed"
                        ) from e
                else:
                    raise RuntimeError(
                        f"Zep batch {batch_id} item submission failed"
                    ) from e

            if len(item_details or []) != len(items):
                recovered_items = self._reconcile_batch_item_count(
                    batch_id,
                    expected_item_count,
                )
                recovered_indexes = {
                    getattr(item, "sequence_index", None)
                    for item in recovered_items
                }
                if (
                    len(recovered_items) == expected_item_count
                    and recovered_indexes == set(range(expected_item_count))
                ):
                    item_details = recovered_items[i:expected_item_count]
                else:

View on GitHub (pinned to b5b53acc57)