feat: Custom Tika-compatible proxy guide for Azure AI Foundry Mistral OCR integration

[linbanana]

Check Existing Issues

• I have searched all existing open AND closed issues and discussions and found none comparable to my request.

Verify Feature Scope

• I believe this feature request is appropriately scoped for the Issues section as described above.

Problem Description

Hi Open WebUI team and community,

I have successfully resolved the RAG content extraction limitations on Windows and complex PDF processing by creating a custom Tika-compatible proxy server that leverages Azure AI Foundry (Mistral OCR).

This setup fully addresses the HTTP PUT binary stream method and the specific JSON Dictionary format expected by Open WebUI's internal loader (avoiding the common AttributeError: 'list' object has no attribute 'get' crash).

I have posted the complete FastAPI proxy implementation script and the configuration guide here:
#19707 (comment)

(Note: Please replace the link above with your actual comment link!)

Hope this guide helps anyone looking to scale their RAG with high-precision enterprise OCR!

Proposed Solution

Proposed Implementation

We can implement a lightweight FastAPI proxy to intercept Open WebUI's file extraction pipeline and bridge it with Azure AI Foundry.

How it works:

1. Handles PUT method: Reads request.body() directly to capture the raw binary file stream sent by Open WebUI.
2. Formats as a flat dictionary: Bypasses Tika's default list-of-dicts response format, returning a standard dictionary with {"X-TIKA:content": extracted_text} to seamlessly integrate with Open WebUI's internal Loader.load() extraction method.

This enables high-precision OCR parsing (including markdown tables and checkboxes) for the RAG ingestion pipeline.

Alternatives Considered

No response

Additional Context

No response

[owui-terminator]

🔍 Related Issues Found

I found some existing issues that might be related. Please check if any of these are duplicates or contain helpful solutions:

1. 🟢 #19707 feat: Mistral OCR additional params
   Both issues are about improving Open WebUI’s OCR/RAG ingestion path with Mistral/Azure OCR integration. The new issue specifically proposes a custom Tika-compatible proxy and notes the limitations of the current Mistral OCR file-transfer flow, which aligns with the broader OCR parameter and endpoint flexibility requested here.
   by theobgbd

2. 🟣 #18531 issue: Compatibility Issues with Newer Tika Versions Break Tika Server Access and OCR Functionality in Non-Docker Deployments
   This issue is directly about Tika server compatibility breaking OCR/content extraction in non-Docker deployments. The new issue also centers on a Tika-compatible proxy and the extraction pipeline, making it closely related to the same integration point and failure mode.
   by AnomeZ · bug

3. 🟣 #21567 issue: Web retrieval cannot load PDF documents
   This issue reports PDF retrieval/extraction failures in the web/RAG pipeline involving Tika, which is the same part of Open WebUI’s ingestion stack affected by the proposed proxy. It’s not Azure-specific, but it is relevant because both involve PDF content extraction failing in the retrieval pipeline.
   by erhhung · bug

4. 🟣 #18101 issue: Tika is unable to extract text from .rtf when it's attached from Windows browser
   This issue is about Tika extraction failing on Windows browser uploads, which matches the new issue’s Windows-focused motivation. Both concern file upload/content extraction quirks in the Tika-based RAG pipeline, even though the file type differs.
   by mkhludnev · bug, good first issue, non-core

---

💡 If your issue is a duplicate, please close it and add any additional details to the existing issue instead.

This comment was generated automatically. React with 👍 if helpful, 👎 if not.

[Classic298]

? What's the feature request?