infiniflow/ragflow · error · ValueError

Invoke JSON argument '{key}' is not JSON-serializable.

Error message

Invoke JSON argument '{key}' is not JSON-serializable.

What it means

Thrown by the Invoke component when validating an argument passed as JSON. Before sending the request, the component runs json.dumps(value, allow_nan=False) on each argument; if the value cannot be serialized (TypeError) or contains NaN/Infinity (ValueError, because allow_nan=False), the argument is rejected. The original exception is chained, and a warning with the key, value type, and repr is logged.

Source

Thrown at agent/component/invoke.py:96

                logging.info(
                    "Invoke JSON arg coercion skipped; value is not valid JSON. key=%s raw=%r error=%s",
                    key,
                    raw_value,
                    exc,
                )
                return raw_value

        try:
            json.dumps(value, allow_nan=False)
        except (TypeError, ValueError) as exc:
            logging.warning(
                "Invoke JSON arg is not JSON-serializable. key=%s value_type=%s value=%r error=%s",
                key,
                type(value).__name__,
                value,
                exc,
            )
            raise ValueError(f"Invoke JSON argument '{key}' is not JSON-serializable.") from exc

        return value

    def get_input_form(self) -> dict[str, dict]:
        res = {}
        for item in self._param.variables or []:
            if not isinstance(item, dict):
                continue
            ref = (item.get("ref") or "").strip()
            if not ref or ref in res:
                continue

            elements = self.get_input_elements_from_text("{" + ref + "}")
            element = elements.get(ref, {})
            res[ref] = {
                "type": "line",
                "name": element.get("name") or item.get("key") or ref,
            }

View on GitHub (pinned to 554fb1133a)

Solutions

1. Convert the offending value to JSON-native types before it reaches Invoke: str()/isoformat() for datetimes, list() for sets, .item() for numpy scalars, dict(model) for ORM rows.
2. Check the preceding log line 'Invoke JSON arg is not JSON-serializable. key=... value_type=... value=...' to identify exactly which key and type failed, then fix that variable's producer.
3. If NaN/Infinity is legitimate in your data, sanitize it explicitly (e.g. replace with None) because the component forbids non-finite floats by design.
4. Add a Code/transform component between the producer and Invoke that normalizes the payload to plain dict/list/str/int/float/bool/None.

Example fix

# before
invoke_args = {"ts": created_at, "tags": {"a", "b"}}  # datetime + set -> raises

# after
invoke_args = {
    "ts": created_at.isoformat() if created_at else None,
    "tags": sorted({"a", "b"}),
}

import json, math

def json_safe(value):
    try:
        json.dumps(value, allow_nan=False)
        return True
    except (TypeError, ValueError):
        return False

# before wiring args into Invoke:
assert all(json_safe(v) for v in invoke_args.values()), invoke_args.keys()

import json

def is_json_serializable(v) -> bool:
    try:
        json.dumps(v, allow_nan=False)
        return True
    except (TypeError, ValueError):
        return False

try:
    invoke_component._invoke(**kwargs)
except ValueError as e:
    if 'not JSON-serializable' in str(e):
        # normalize the offending variable and retry once
        ...

When it happens

Trigger: Calling an Invoke component whose variables resolve to a non-JSON value: a datetime/date object, set, bytes, numpy scalar, or an ORM/model instance; or a float that is NaN/Inf (allow_nan=False makes even json.dumps itself raise ValueError). The failure occurs at argument-validation time, before any HTTP request is made.

Common situations: Upstream component (e.g. Code, Iteration, LLM returning structured output parsed into Python objects) produces datetime or numpy types that flow into an Invoke argument; dividing float values that yield NaN; passing a Peewee model row instead of a plain dict.

Related errors

• Invoke headers must be a JSON object.
• flow.formatTypeError
• Login failed: invalid JSON response ({exc})
• {} must be an array, but its type is {}
• The input of List Operations should be an array.

AI-assisted analysis of infiniflow/ragflow@554fb1133a (2026-08-15). Data as JSON: /api/errors/c3393954dd1059bd. Report an issue: GitHub.