Upload as Text not sending the user message along with the system message

[dominicfallows]

I'm using Upload as Text and a custom endpoint. The user message, or text as the LibreChat UI to LibreChat Agents API has it, isn't being forwarded on to the custom endpoint. It is only sending the system message, or ocr in the initial UI to Agents API call has it.

Is this expected? I ask as the LLM is receiving the OCR content, but without any instruction as to what to do with it.

Is there any config to enable the user message to come through with the first upload?

Here's some debug logs:

LibreChat logs

debug: [BaseClient] userMessage
{
  messageId: "11d96447-3f1b-48ec-b085-972e4ab43335",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
  conversationId: "5d962b06-213b-452a-aaf2-98970ea25b7b",
  sender: "User",
  text: "Analyse this checklist and suggest improvements, clarifications, or simplifications I can make to ma... [truncated]",
  isCreatedByUser: true,
  image_urls: undefined,
  ocr: "Attached document(s):\n```md# \"218548137_51bb0e2bbe664f4db29fa1b843b8106b-031025-1111-908.pdf\"\n# PAGE... [truncated]",
  tokenCount: 32,
}

and

sending request {
  method: 'post',
  url: 'http://router:8000/v1/chat/completions',
  options: {
    method: 'post',
    path: '/chat/completions',
    body: {
      model: 'auto',
      temperature: undefined,
      top_p: undefined,
      frequency_penalty: undefined,
      presence_penalty: undefined,
      logprobs: undefined,
      top_logprobs: undefined,
      n: undefined,
      logit_bias: undefined,
      stop: undefined,
      user: '[truncated]',
      stream: true,
      functions: undefined,
      function_call: undefined,
      tools: undefined,
      tool_choice: undefined,
      response_format: undefined,
      seed: undefined,
      parallel_tool_calls: undefined,
      max_tokens: undefined,
      messages: [
        {
            "role": "system",
            "content": "Attached document(s):\n```md# \"218548137_51bb0e2bbe664f4db29fa1b843b8106b-031025-1111-908.pdf\"\n# PAGE 1\n# ...[truncated]```"
         }
    ]
    },
    organization: undefined,
    dangerouslyAllowBrowser: true,
    baseURL: 'http://router:8000/v1',
    signal: AbortSignal { aborted: false },
    stream: true
  },
  ...[truncated]
}

My Custom Endpoint receiving logs

DEBUG] /v1/chat/completions
{
    "model": "auto",
    "user": "[redacted]",
    "stream": true,
    "messages": [
        {
            "role": "system",
            "content": "Attached document(s):\n```md# \"218548137_51bb0e2bbe664f4db29fa1b843b8106b-031025-1111-908.pdf\"\n# PAGE 1\n# ...[truncated]```"
         }
    ]
}

Extract of my librechat.yaml

ocr:
  mistralModel: "mistral-document-ai-2505-uksouth"
  apiKey: "${AZURE_MISTRAL_OCR_API_KEY}"
  baseURL: "https://[redacted].services.ai.azure.com/providers/mistral/azure"
  strategy: "azure_mistral_ocr"
...
fileConfig:
  ...
  # Files matching these patterns use OCR (if configured)
  ocr:
    supportedMimeTypes:
      - "^image/(jpeg|gif|png|webp|heic|heif)$"
      - "^application/pdf$"
      - "^application/vnd\\.openxmlformats-officedocument\\.(wordprocessingml\\.document|presentationml\\.presentation|spreadsheetml\\.sheet)$"
      - "^application/vnd\\.ms-(word|powerpoint|excel)$"
      - "^application/epub\\+zip$"

  # Files matching these patterns use text parsing
  text:
    supportedMimeTypes:
      - "^text/(plain|markdown|csv|json|xml|html|css|javascript|typescript|x-python|x-java|x-csharp|x-php|x-ruby|x-go|x-rust|x-kotlin|x-swift|x-scala|x-perl|x-lua|x-shell|x-sql|x-yaml|x-toml)$"
...
endpoints:
    agents:
        capabilities:
            - file_search 
            - artifacts
            - context
            - ocr
            - chain
        ...

[dominicfallows]

Update: I'm closing this down as I found the resolution. In my librechat.yaml for my endpoint I did not have maxContextTokens set. Whatever the default value is for this was causing the User message part to be excluded from the request to my endpoint. By setting a value (FYI, I chose 16k) the User message part was included.