500 Internal Server Error — “no messages in fragment” when using /v1/responses endpoint Image Input

When sending a POST request to the /v1/responses endpoint with the parameter input, the API returns a 500 error with the message:

{"error":"no messages in fragment"}

500
{"id":"","object":"","created_at":0,"status":"","error":"no messages in fragment","incomplete_details":null,"instructions":null,"max_output_tokens":null,"model":"","output":null,"parallel_tool_calls":false,"previous_response_id":null,"reasoning":{},"store":false,"temperature":0,"text":{},"tool_choice":"","tools":null,"top_p":0,"truncation":"","usage":{"input_tokens":0,"input_tokens_details":{"cached_tokens":0},"output_tokens":0,"output_tokens_details":{"cached_tokens":0},"total_tokens":0},"user":null,"metadata":null}

Steps to Reproduce

1. Send a request similar to the following:

import requests, json

url = "http://localhost:8080/v1/responses"
headers = {"Content-Type": "application/json"}
data = {
    "model": "LocalAI_Agent",
    "input": [
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "what is in this image?"},
                {
                    "type": "input_image",
                    "image_url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAMsAAAD4CAMAAAB1y+ICAAAB6VBMVEAFADpnqTrj4znsqTZb3kfTWF7CCC3CiuAAAkAFRY9PUcAVmIAJS1/ZA2xkQVENwLPrg1gUwqzmgpeBh6YABmZfgtHMwhmVwWdfgN7cAl3aCr73QEeVnLtUXp5oprleXrnuVZDCZTINtydpLvb0tidZfG9H28RZi7FeL5H/otPT/AdHKBBs1r1ZSAAAAAElFTkSuQmCC..."
                }
            ]
        }
    ]
}

response = requests.post(url, headers=headers, data=json.dumps(data))
print(response.status_code, response.text)

Expected Behavior

Should return a valid model response analyzing the text and image input.

Actual Behavior

Returns:

500 Internal Server Error
{"error":"no messages in fragment"}

Possible Cause

The /v1/responses endpoint appears to require a messages array instead of an input array, following the newer message-based schema. When using input, the request payload does not match the expected structure, causing the backend to throw an internal error instead of returning a user-facing validation message.

Suggested Fix

• Validate incoming requests for expected fields (messages vs input) and return a 400 Bad Request with a descriptive message rather than 500.
• Update documentation to clarify the correct parameter format for multimodal input.

Start at the /v1/responses endpoint and reproduce the supplied payload with the input array and image content. Determine whether input or messages is the supported schema, then verify that malformed requests receive a descriptive 400 response rather than the current 500; clarify the multimodal format in the documentation if needed.