langflow-ai/langflow · error · ValueError

Error serializing vertex build response: {exc}

Error message

Error serializing vertex build response: {exc}

What it means

A ValueError raised when a vertex build succeeded but its VertexBuildResponse cannot be serialized with model_dump_json()/json.loads() before emitting the on_end_vertex event. This indicates the response model contains data that pydantic cannot dump to JSON — typically non-serializable objects placed in the output data by a custom component.

Source

Thrown at src/backend/base/langflow/api/build.py:831

        # Why: the background path never enters Graph.process(), so the pause boundary must live in this driver.
        await graph.check_and_handle_pause()
        try:
            vertex_build_response: VertexBuildResponse = await _build_vertex(vertex_id, graph, event_manager)
        except asyncio.CancelledError:
            await logger.ainfo("Build cancelled")
            raise

        # Accumulate the vertex timedelta
        if vertex_build_response.data.timedelta is not None:
            vertex_timedeltas.append(vertex_build_response.data.timedelta)

        # send built event or error event
        try:
            vertex_build_response_json = vertex_build_response.model_dump_json()
            build_data = json.loads(vertex_build_response_json)
        except Exception as exc:
            msg = f"Error serializing vertex build response: {exc}"
            raise ValueError(msg) from exc

        event_manager.on_end_vertex(
            data={"build_data": build_data, "output_meta": _output_meta_for_vertex(graph, vertex_id)}
        )

        if vertex_build_response.valid and vertex_build_response.next_vertices_ids:
            tasks = []
            for next_vertex_id in vertex_build_response.next_vertices_ids:
                task = asyncio.create_task(
                    build_vertices(
                        next_vertex_id,
                        graph,
                        event_manager,
                        vertex_timedeltas,
                    )
                )
                tasks.append(task)
            await asyncio.gather(*tasks)

View on GitHub (pinned to 976ec789d2)

Solutions

1. In the failing component, convert the output to a supported type (Message, Data, str, or a plain dict/list of primitives) before returning
2. If returning rich objects, implement model_serializer / dict() conversion, or use repr()/str() for display outputs
3. Pin consistent pydantic versions across the workspace if serialization behavior changed after upgrade
4. Reproduce by building only that vertex to confirm the serialization fix

Example fix

# before
return Data(data={'result': some_llm_sdk_object})  # not JSON-dumpable
# after
return Data(data={'result': str(some_llm_sdk_object)})
# or the idiomatic way:
return Message(text=some_llm_sdk_object.content)

# before returning from a custom component, ensure JSON-serializability
import json
def safe_output(obj) -> str:
    try:
        json.dumps(obj)
        return obj
    except TypeError:
        return str(obj)

from langflow.schema.message import Message
from langflow.schema import Data
from langflow.schema.dataframe import DataFrame

def is_serializable_output(v) -> bool:
    return isinstance(v, (Message, Data, DataFrame, str, int, float, bool, list, dict, type(None)))

try { ... } finally { /* vertex serialization failure is a server bug-class error: fix the component rather than catching at the edge */ }

When it happens

Trigger: A component returning raw objects (DB cursors, tensors, file handles, arbitrary class instances) in its Output data instead of langflow Message/Data types; outputs whose type fails pydantic-core serialization (e.g. objects with __slots__ cycles, numpy types without serializers).

Common situations: Custom components returning SDK objects directly instead of calling .data or converting to str/Message; library upgrades changing pydantic serialization strictness.

Related errors

• HTTP error! status: ${response.status}
• Could not extract class name from code
• Inputs and outputs have overlapping names: {overlap}
• {e}
• actions must be strings

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/c96a1817c4ac8f7a. Report an issue: GitHub.