ollama: attaching a file does not work/results in error 500/400 Bad request

[nodiscc]

Steps to reproduce:

• llms --serve 8080 --verbose, access web UI
• select ollama provider (tested with multiple models)
• attach a file (tested with multiple text files/shell/python scripts) with a promt su chas Please summarize this document and submit the query
• the web UI displays an error and not response is returned

HTTP 500 Internal Server Error
{
  "error": "HTTP 400 Bad Request"
}

Logs:

provider: ollama OllamaProvider
POST http://localhost:11434/v1/chat/completions
{
  "model": "deepseek-r1:14b",
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "file",
          "file": {
            "filename": "test.xml",
            "file_data": "data:text/xml;base64,(1325)"
          }
        },
        {
          "type": "text",
          "text": "Summarize the file\n\n[\ud83d\udcce test.xml]"
        }
      ]
    }
  ],
  "enable_thinking": true,
  "metadata": {
    "threadId": "1765739448549"
  },
  "stream": false
}
Provider ollama failed: HTTP 400 Bad Request

[nodiscc]

Interestingly, attaching an image file and asking the ollama model to describe it works as expected.
I see it uses "type": "image_url" instead of "type": "file"

Details

provider: ollama OllamaProvider
Resized image from 1754x1403 to 1280x1024
Converted image to WebP: 798373 bytes -> 235210 bytes (29%)
POST http://localhost:11434/v1/chat/completions
{
  "model": "qwen3-vl:8b",
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/webp;base64,(313617)"
          }
        },
        {
          "type": "text",
          "text": "Describe the key features of the input image\n\n[\ud83d\uddbc\ufe0f Snowden_Slights_with_big_gun_YORYM-S364.jpg]"
        }
      ]
    }
  ],
  "enable_thinking": true,
  "metadata": {
    "threadId": "1765745070478"
  },
  "stream": false
}
{
  "id": "chatcmpl-198",
  "object": "chat.completion",
  "created": 1765745083,
  "model": "qwen3-vl:8b",
  "system_fingerprint": "fp_ollama",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "This black - and - white photograph presents a historical scene with several prominent features:  [...].",
        "reasoning": "So, let's describe the key features of this black-and-white photograph. [...]."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 1328,
    "completion_tokens": 687,
    "total_tokens": 2015
  },
  "metadata": {
    "duration": 12785,
    "pricing": "0/0"
  }
}

[nodiscc]

      "content": [
        {
          "type": "file",
          "file": {
            "filename": "test.xml",
            "file_data": "data:text/xml;base64,(1325)"
          }

I don't think this is supposed to be supported at all by the API (https://docs.ollama.com/api/openai-compatibility), am I correct? This issue seems to support my conclusion ollama/ollama#5762

I think the only correct way to achieve this, would be to have llms.py extract the content of the file and pass it to the LLM through the prompt.

[mythz]

I can't find any official way that Ollama supports to send file attachments like PDFs, the standard OpenAI compatible way works if I call the same model on OpenRouter (e.g. Gemma 3n 4B) so I don't think it's supported.

I'll wait for official support from Ollama, which I expect will just be to implement OpenAI's file type support, like they do with image_url when they added support for it.