langflow-ai/langflow · error · HTTPException

Flow module must define 'get_graph()' function: {flow_path}

Error message

Flow module must define 'get_graph()' function: {flow_path}

What it means

Raised when flow_loader dynamically imports a Python flow module (.py file) but the module exposes neither a get_graph() function nor a module-level graph variable. Langflow requires one of these as the entry point to materialize a runnable graph. It surfaces as HTTPException 500 from the flow-loading endpoint.

Source

Thrown at src/backend/base/langflow/agentic/services/helpers/flow_loader.py:161

        with _temporary_sys_path(str(flow_path.parent)):
            spec.loader.exec_module(module)
    except Exception as e:
        if module_name in sys.modules:
            del sys.modules[module_name]
        logger.error(f"Error loading Python flow module: {e}")
        raise HTTPException(status_code=500, detail=f"Error loading flow module: {e}") from e

    if not hasattr(module, "get_graph"):
        # Fallback: check for 'graph' variable for backward compatibility
        if hasattr(module, "graph"):
            graph = module.graph
            validate_flow_for_current_settings(graph)
            if module_name in sys.modules:
                del sys.modules[module_name]
            return graph
        if module_name in sys.modules:
            del sys.modules[module_name]
        raise HTTPException(status_code=500, detail=f"Flow module must define 'get_graph()' function: {flow_path}")

    get_graph_func = module.get_graph

    # Build kwargs for get_graph based on what it accepts
    sig = inspect.signature(get_graph_func)
    kwargs = {}
    if "provider" in sig.parameters and provider:
        kwargs["provider"] = provider
    if "model_name" in sig.parameters and model_name:
        kwargs["model_name"] = model_name
    if "api_key_var" in sig.parameters and api_key_var:
        kwargs["api_key_var"] = api_key_var
    # Python flows never pass through the JSON-side inject_iterations_into_flow,
    # so the runtime step budget must be forwarded to get_graph explicitly.
    raw_iterations = (provider_vars or {}).get("ITERATIONS_LIMIT")
    if "iterations_limit" in sig.parameters and raw_iterations not in (None, ""):
        with suppress(TypeError, ValueError):
            kwargs["iterations_limit"] = int(raw_iterations)

View on GitHub (pinned to 976ec789d2)

Solutions

1. Add a module-level def get_graph() -> Graph: that builds and returns the flow's Graph object.
2. Alternatively define a module-level graph variable for backward compatibility (the loader accepts it as a fallback).
3. Check for a typo: the loader matches the exact names 'get_graph' and 'graph'.
4. Verify you pointed flow_path at the entry file of the flow, not a helper module.

Example fix

# before
# my_flow.py
def build_graph():
    return Graph(...)

# after
# my_flow.py
def get_graph():
    return Graph(...)

import importlib.util

def flow_module_is_loadable(flow_path: str) -> bool:
    spec = importlib.util.spec_from_file_location("_probe", flow_path)
    if spec is None or spec.loader is None:
        return False
    mod = importlib.util.module_from_spec(spec)
    try:
        spec.loader.exec_module(mod)
    except Exception:
        return False
    return hasattr(mod, "get_graph") or hasattr(mod, "graph")

def has_graph_entrypoint(module) -> TypeGuard[Any]:
    return hasattr(module, "get_graph") or hasattr(module, "graph")

try:
    graph = load_flow_module(flow_path)
except HTTPException as e:
    if "must define 'get_graph()'" in e.detail:
        raise ValueError(f"{flow_path} lacks get_graph()/graph") from e
    raise

When it happens

Trigger: Calling the agentic flow-loader endpoint with a flow_path pointing at a Python file whose module defines neither 'get_graph' nor 'graph' (e.g. only helper functions, or the graph factory is named differently like build_graph()). The module is imported, inspected with hasattr for both symbols, and rejected when both checks fail.

Common situations: Migrating flows written for an older loader that expected a different factory name; typos in the function name (getGraph, get_graph_async); a module that conditionally defines get_graph only under __main__; picking the wrong file out of a multi-file flow project.

Related errors

• Error creating graph: {e}
• Invalid flow filename
• Invalid flow filename: '{flow_filename}'
• Flow file '{flow_filename}' not found
• Could not load flow module: {flow_path}

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/e4522b0f659d2d4f. Report an issue: GitHub.