litellm - 💡(How to fix) Fix [Bug]: Chat completions → Responses API bridge does not handle file_id with URL (documented LiteLLM pattern) [1 participants]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

LiteLLM's document understanding documentation explicitly shows that passing an HTTPS URL in file_id is the supported way to send a file by URL through the Chat Completions API:

file_url = "https://www.w3.org/WAI/ER/tests/xhtml/testfiles/resources/pdf/dummy.pdf"
file_content = [
    {"type": "text", "text": "What's this file about?"},
    {"type": "file", "file": {"file_id": file_url, "format": "application/pdf"}}
]
response = completion(model="...", messages=[{"role": "user", "content": file_content}])

This pattern works correctly for Gemini, Bedrock, and Anthropic bridges. However, when the model routes through the chat completions → Responses API bridge (i.e. a model with mode: "responses"), the bridge passes the URL through as input_file.file_id unchanged, which the Responses API rejects:

{
  "error": {
    "message": "Invalid 'input[0].content[1].file_id': 'https://arxiv.org/pdf/1706.03762'. Expected an ID that contains letters, numbers, underscores, or dashes, but this value contained additional characters.",
    "type": "invalid_request_error",
    "param": "input[0].content[1].file_id",
    "code": "invalid_value"
  }
}

The correct behaviour is: when file_id contains an HTTPS URL, the bridge should map it to file_url in the Responses API input_file, since file_url is the Responses API's native field for URL-based file input.

Steps to Reproduce

1. Register an Azure model with mode: "responses":

curl -X POST "http://localhost:4000/model/new" \
  -H "Authorization: Bearer <master_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model_name": "my-responses-model",
    "litellm_params": {
      "model": "azure/gpt-4.1-mini",
      "api_base": "https://<account>.openai.azure.com/openai/v1"
    },
    "model_info": {"mode": "responses"}
  }'

1. Send a chat completions request using the documented file_id-with-URL pattern:

curl -X POST "http://localhost:4000/v1/chat/completions" \
  -H "Authorization: Bearer <key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "my-responses-model",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "What is this document about?"},
        {"type": "file", "file": {"file_id": "https://arxiv.org/pdf/1706.03762", "format": "application/pdf"}}
      ]
    }]
  }'

1. Observe the error — the URL is passed through as input_file.file_id, which the Responses API rejects.

Expected behaviour

The bridge should detect that file_id contains an HTTPS URL and map it to input_file.file_url in the Responses API request, consistent with how the Gemini bridge handles the same pattern.

Root cause

In litellm/completion_extras/litellm_responses_transformation/transformation.py, the file conversion (lines ~785-790) copies file_id from the chat completion message directly into the Responses API input_file without checking whether the value is a URL:

converted = {"type": "input_file"}
if isinstance(file_data, dict):
    for key in ["file_id", "file_data", "filename"]:
        if key in file_data:
            converted[key] = file_data[key]

The fix would be: if file_id starts with https://, use file_url instead of file_id in the converted output.

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on?

main (verified against current main branch code)

Twitter / LinkedIn details

No response