DeepSeek and Grok (xAI) configuration: File upload, web search limitations, and required workarounds

[fabbrito]

There are several discussions in this repository regarding file upload features and usability enhancements (such as in #10560). However, my situation is a bit different: I want to focus on the limitations and configuration needs for using DeepSeek and Grok (xAI) models specifically, especially when integrating file upload, OCR, and web search within LibreChat.

My Experience

• Prior to v0.7.8 (I think), uploading files was supported through LibreChat’s built-in pdf/docx/OCR backend, even when the model itself (like DeepSeek or Grok) doesn’t support native file uploads.
• Since then, sending files directly to DeepSeek or Grok leads to errors, since neither truly supports file uploads, and the default seamless fallback is gone.
• Additionally, both DeepSeek and Grok (xAI) do not support web search/browse tools, but those are enabled by default.

My Current Configuration

Configured librechat.yaml:

endpoints:
  custom:
    - name: "Deepseek"
      apiKey: "${DEEPSEEK_API_KEY}"
      baseURL: "https://api.deepseek.com/v1"
      models:
        default: ["deepseek-chat", "deepseek-reasoner"]
        fetch: false
      titleConvo: true
      titleModel: "deepseek-chat"
      modelDisplayLabel: "Deepseek"

My Grok configuration is similar, adjusted for the correct endpoint and model names.

Current Workarounds

LibreChat offers two strategies for file upload:

1. Sending files directly to the LLM (if supported)
2. Using an OCR reader (like Mistral OCR) to extract text and send the text to the LLM

For DeepSeek and Grok, only the OCR approach works. I process files via OCR, then send the resulting text prompt to the model. I must also disable resendFiles in the model parameters, because enabling it with DeepSeek/Grok results in errors or no output.

Furthermore, because neither DeepSeek nor Grok supports web search, I created a custom ModelSpec/preset that disables web search and browsing capabilities. Without this, attempts to use these features degrade the experience or fail.

Open Questions / Pains

• Is this approach (OCR for files, disabling resendFiles, and disabling web search) the best recommended practice for users of DeepSeek and Grok in LibreChat?
• Are improvements planned (or possible) for automatic detection/disablement of unsupported features for endpoints/models?

Request

• I’d appreciate any advice about practical configuration and preset setup for DeepSeek and Grok in LibreChat, specifically relating to file upload, OCR, and tool integration.
• Please share workarounds, known issues, or upcoming plans for simplifying this experience for these models.
• If you know of other relevant file-upload/LLM integration discussions, please mention them here as well!

Thanks for any suggestions or official guidance!