function_openai_responses.ResponsesBody() got multiple values for keyword argument 'user'

[pascal-sidlee]

Hi,

On the main branch.

If we use a custom Filter who adds a "user" in body, when the pipeline openai_responses_manifold.pyis used, we get this error: function_openai_responses.ResponsesBody() got multiple values for keyword argument 'user'.

If we add this check, everything is working succesfully:

# Start at Line 653

# Transform request body (Completions API -> Responses API).
completions_body = CompletionsBody.model_validate(body)

# Check if 'user' already exists in the body to avoid duplication
has_user = hasattr(completions_body, "user") or "user" in body

responses_body = ResponsesBody.from_completions(
    completions_body=completions_body,
    # If chat_id and openwebui_model_id are provided, from_completions() uses them to fetch previously persisted items (function_calls, reasoning, etc.) from DB and reconstruct the input array in the correct order.
    **(
        {"chat_id": __metadata__["chat_id"]}
        if __metadata__.get("chat_id")
        else {}
    ),
    **(
        {"openwebui_model_id": openwebui_model_id} if openwebui_model_id else {}
    ),
    # Additional optional parameters passed directly to ResponsesBody without validation. Overrides any parameters in the original body with the same name.
    truncation=valves.TRUNCATION,
    **({"user": user_identifier} if not has_user else {}),  # Only add user if not already present
    service_tier=valves.SERVICE_TIER,
    **(
        {"max_tool_calls": valves.MAX_TOOL_CALLS}
        if valves.MAX_TOOL_CALLS is not None
        else {}
    ),
)

[jrkropp]

@pascal-sidlee If you are just trying to add the Open WebUI user, you can enable it via the Manifold's valve. Set 'Prompt Cache Key' to email. Let me know if you plan to inject something else and I can make the adjustment.

[pascal-sidlee]

Basically, just before calling this pipeline, I have a Filter who add user into body. Then after the pipeline failed, because of an existing user attribute.

Here is my Filter code:

from typing import Any, Dict, Optional

try:
    from pydantic import BaseModel  # Optional dependency
except ImportError:  # pragma: no cover
    BaseModel = object


class Filter:
    class Valves(BaseModel):
        # Future configuration knobs—for now this is intentionally blank.
        pass

    def __init__(self) -> None:
        self.valves = self.Valves()

    # ---------- Inlet -------------------------------------------------
    def inlet(
        self,
        body: Dict[str, Any],
        __user__: Optional[Dict[str, Any]] = None,
    ) -> Dict[str, Any]:
        """
        Attach the current user's ID to the outgoing LLM payload.

        Follows OpenAI's recommendation of a <= 64-character string.
        """
        if __user__ is None or "user" in body:
            return body  # nothing to do

        user_id = __user__.get("id")
        if not isinstance(user_id, (str, int)):
            # Optionally: log.debug("Invalid user ID: %r", user_id)
            return body

        user_id_str = str(user_id)
        if len(user_id_str) > 64:  # OpenAI hard limit
            return body

        new_body = dict(body)  # avoid mutating shared dict
        new_body["user"] = user_id_str # <-- this line break your pipeline
        return new_body

[jrkropp]

Basically, just before calling this pipeline, I have a Filter who add user into body. Then after the pipeline failed, because of an existing user attribute.

Here is my Filter code:

from typing import Any, Dict, Optional

try:
from pydantic import BaseModel # Optional dependency
except ImportError: # pragma: no cover
BaseModel = object

class Filter:
class Valves(BaseModel):
# Future configuration knobs—for now this is intentionally blank.
pass

def __init__(self) -> None:
    self.valves = self.Valves()

# ---------- Inlet -------------------------------------------------
def inlet(
    self,
    body: Dict[str, Any],
    __user__: Optional[Dict[str, Any]] = None,
) -> Dict[str, Any]:
    """
    Attach the current user's ID to the outgoing LLM payload.

    Follows OpenAI's recommendation of a <= 64-character string.
    """
    if __user__ is None or "user" in body:
        return body  # nothing to do

    user_id = __user__.get("id")
    if not isinstance(user_id, (str, int)):
        # Optionally: log.debug("Invalid user ID: %r", user_id)
        return body

    user_id_str = str(user_id)
    if len(user_id_str) > 64:  # OpenAI hard limit
        return body

    new_body = dict(body)  # avoid mutating shared dict
    new_body["user"] = user_id_str # <-- this line break your pipeline
    return new_body

The pipe already injects either the Open WebUI's userid OR email automatially. So you could just disable the filter and use the built in functionality.