PDF Uploads Broken with Google Gemini via Vertex AI - "Error Processing File" - ALL Gemini Models Affected

[marlonka]

What happened?

In the latest release (v0.7.7-rc1, as seen in the screenshot), PDF uploads are consistently failing when using any Google Gemini model via Vertex AI. Every attempt to upload a PDF results in an "Error Processing file" message being displayed. Once a PDF has been added to a conversation, the conversation also can not be continued, as the send button is greyed out.

Expected behavior

The PDF file should be processed successfully, allowing the assistant to access and utilize the PDF's content.

Environment:

• LibreChat Version: v0.7.7-rc1 (as seen in the screenshot)
• Model: Google Gemini (all versions tested via Vertex AI. Example: gemini-2.0-flash-001, as seen in the screenshot)
• Provider: Vertex AI
• Install Type: Fresh install from the LibreChat GitHub repository on 2025-02-13 (February 13, 2025).

Possible Causes

• Possible incompatibility between the current PDF processing logic and the Gemini API (across all Gemini models).
• Issue with authentication or API keys related to Vertex AI (affecting Gemini integration).
• A change in the PDF processing library or method used in the latest LibreChat release (that disproportionately affects Gemini).
• Potentially an issue in the Gemini integration code itself within LibreChat.

Thank you for looking into this! This is a complete blocker for us

Version Information

• LibreChat Version: v0.7.7-rc1 (as seen in the screenshot)

Steps to Reproduce

1. Configure LibreChat to use any Google Gemini model via Vertex AI.
2. Attempt to upload any PDF file.
3. Observe the "Error Processing file" message appear.

What browsers are you seeing the problem on?

Chrome

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[hamletfg]

Tested PDF Uploads on Google AI studio API and it works, so it seems isolated just to Vertex AI models.

[danny-avila]

Is your RAG setup configured? Otherwise, direct uploads to provider are not yet supported, please track: #2755

[marlonka]

I just copied env.example and this was a fresh install. I thought it should work at least. what do i need to change?

[danny-avila]

it works with RAG configured, please see the docs

[marlonka]

sorry danny, it works with OpenAI as RAG embedding model configured. (have to check whether it works with Azure OpenAI too, that was the original problem, because Vertex AI didnt work even though RAG API with azure openai was configured)

[marlonka]

just for you to know the mistake was not RAG API actually but we had PDF_EXTRACT_IMAGES: true but not updated RAG API to dev latest that was the issue