langchain-ai/langchain · error · ValueError

Invalid format: {self._schema_format}

Error message

Invalid format: {self._schema_format}

What it means

Raised by BaseLangChainTracer._get_chain_inputs when normalizing chain run inputs for a tracer whose _schema_format is not one of the supported values ('original', 'original+chat', 'streaming_events'). The tracer cannot decide whether to wrap non-dict inputs under an 'input' key or pass them through, so it refuses to record the run. It indicates a misconstructed or version-mismatched tracer rather than bad user data.

Source

Thrown at libs/core/langchain_core/tracers/core.py:387

            extra=kwargs,
            events=[{"name": "start", "time": start_time}],
            start_time=start_time,
            child_runs=[],
            run_type=run_type or "chain",
            name=name,
            tags=tags or [],
        )

    def _get_chain_inputs(self, inputs: Any) -> Any:
        """Get the inputs for a chain run."""
        if self._schema_format in {"original", "original+chat"}:
            return inputs if isinstance(inputs, dict) else {"input": inputs}
        if self._schema_format == "streaming_events":
            return {
                "input": inputs,
            }
        msg = f"Invalid format: {self._schema_format}"
        raise ValueError(msg)

    def _get_chain_outputs(self, outputs: Any) -> Any:
        """Get the outputs for a chain run."""
        if self._schema_format in {"original", "original+chat"}:
            return outputs if isinstance(outputs, dict) else {"output": outputs}
        if self._schema_format == "streaming_events":
            return {
                "output": outputs,
            }
        msg = f"Invalid format: {self._schema_format}"
        raise ValueError(msg)

    def _complete_chain_run(
        self,
        outputs: dict[str, Any],
        run_id: UUID,
        inputs: dict[str, Any] | None = None,
    ) -> Run:

View on GitHub (pinned to e32fa9a52e)

Solutions

1. Do not set _schema_format yourself; use the default ('original') or the documented values 'original+chat' / 'streaming_events'
2. If subclassing, validate the value in __init__ against {'original','original+chat','streaming_events'} and raise early
3. Align versions: uv sync so every langchain-* package uses the same langchain-core version
4. If you need a custom input schema, override _get_chain_inputs/_get_chain_outputs entirely instead of inventing a format string

Example fix

// before
class MyTracer(BaseLangChainTracer):
    def __init__(self):
        self._schema_format = "my_format"
// after
class MyTracer(BaseLangChainTracer):
    def __init__(self):
        self._schema_format = "original"  # or 'streaming_events'

VALID_FORMATS = {"original", "original+chat", "streaming_events"}
assert tracer._schema_format in VALID_FORMATS, tracer._schema_format

def is_valid_schema_format(fmt: str) -> bool:
    return fmt in {"original", "original+chat", "streaming_events"}

When it happens

Trigger: Instantiating BaseLangChainTracer (or a subclass) and setting _schema_format to an arbitrary/custom string; passing a tracer built against a newer langchain-core (which introduced new formats) into an older runtime that does not recognize it; calling on_chain_start with such a tracer attached to config['callbacks'].

Common situations: Custom tracer subclasses that override __init__ and set _schema_format from a constructor kwarg without validating it; mixing langchain-core versions in one environment (e.g. a partner package pinning an older core); copy-pasted tracer code from a different core version.

Related errors

• Tracing using LangChainTracerV1 is no longer supported. Plea
• Could not resolve content_key {full_path!r}: expected a mapp
• Could not resolve content_key {full_path!r}: missing key {ke
• If env_var is set, handle_class must also be set to a non-No
• A pending deprecation cannot have a scheduled removal

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/e25aa7faa89e48b9. Report an issue: GitHub.