feat: Option to set "full document mode" in uploaded documents as default. #6754

New Issue

GiteaMirror · 2025-11-11T17:05:07-06:00

GiteaMirror commented

2025-11-11 17:05:07 -06:00

Originally created by @deliciousbob on GitHub (Oct 24, 2025).

Check Existing Issues

I have searched all existing open AND closed issues and discussions for similar requests. I have found none that is comparable to my request.

Problem Description

We get most complains from our users:

Document upload and questioning against documents
(summarize this, what is xy about, create following tasks, ...) often ends in:
- eather not document related related response (general llm knowledge)
- or llm mentions "cannot find the document", "there is no document xy".

We then tell the users to click into the document and toggle on "full context mode".
This helped in most cases to get a good response related to the document.

Most documents are relative small pdfs, so a RAG mode is not needed or contraproductive.

Desired Solution you'd like

Good Day Developers,
It would be great to have a Env. Variable or a toggle in the Admin-Settings to force users to use "full context / use entire document" mode for uploaded documents. (but they can change to focused retrieval mode manually - so the other way around)

Thank you very much!
Best regards, Robert

Alternatives Considered

Even better would be to be able to set a Token Limit per Document after that it uses the focus retrieval mode automatically, everything below would use the "full context / entire document" mode.

Document is uploaded
1.2 text extraction process
1.3 Count tokens from context
IF context < token.limit --> enable full document mode
Else --> focused retrieval mode

Additional Context

Users tend to compare features to OpenAI's Chat plattform, they can simply upload files and ask agains these files.
That works very well, with OpenWebUI it works well if you set the full document mode.

Create a Summary for example, does not work in retrieval mode.

Originally created by @deliciousbob on GitHub (Oct 24, 2025). ### Check Existing Issues - [x] I have searched all existing open AND closed issues and discussions for similar requests. I have found none that is comparable to my request. ### Problem Description We get most complains from our users: - Document upload and questioning against documents (summarize this, what is xy about, create following tasks, ...) often ends in: - eather not document related related response (general llm knowledge) - or llm mentions "cannot find the document", "there is no document xy". We then tell the users to click into the document and toggle on "full context mode". This helped in most cases to get a good response related to the document. Most documents are relative small pdfs, so a RAG mode is not needed or contraproductive. ### Desired Solution you'd like Good Day Developers, It would be great to have a Env. Variable or a toggle in the Admin-Settings to force users to use "full context / use entire document" mode for uploaded documents. (but they can change to focused retrieval mode manually - so the other way around) <img width="150" height="90" alt="Image" src="https://github.com/user-attachments/assets/26be337a-0824-426b-acc4-320d11595be7" /> Thank you very much! Best regards, Robert ### Alternatives Considered Even better would be to be able to **set a Token Limit** per **Document** after that it uses the focus retrieval mode automatically, everything below would use the "full context / entire document" mode. 1. Document is uploaded 1.2 text extraction process 1.3 Count tokens from context 2. IF context < token.limit --> enable full document mode Else --> focused retrieval mode ### Additional Context Users tend to compare features to OpenAI's Chat plattform, they can simply upload files and ask agains these files. That works very well, with OpenWebUI it works well if you set the full document mode. Create a Summary for example, does not work in retrieval mode.

GiteaMirror closed this issue

2025-11-11 17:05:08 -06:00

GiteaMirror commented

2025-11-11 17:05:09 -06:00

@Classic298 commented on GitHub (Oct 24, 2025):

Possible in admin settings. Enable "bypass embedding and retrieval".

Please use the search

Duplicate

@Classic298 commented on GitHub (Oct 24, 2025): Possible in admin settings. Enable "bypass embedding and retrieval". Please use the search Duplicate

GiteaMirror commented

2025-11-11 17:05:09 -06:00

@deliciousbob commented on GitHub (Oct 24, 2025):

Hi @Classic298, sry I forgot to mention that we have tried that option "bypass embedding and retrieval" at Document settings already, but this seems to do nothing to the file upload. There is still the focus retrieval process active.

And it seems to just disable reranking, what we basically require for Knowledge collection retrieval.
I would love to have everything towards Knowledge-Collections as it is currently with reranking as this works best.

But for single Document uploading (small documents with just some pages) the retrieval process is working worse then
full document mode. So a possibility to switch that behavior to make full document mode default (by env. variable) would be a great addition for the project.

The token limit option that i mentioned in the Alternatives Considered section on top would be the cherry on top.
Thx, best regards, Robert

@deliciousbob commented on GitHub (Oct 24, 2025): Hi @Classic298, sry I forgot to mention that we have tried that option "bypass embedding and retrieval" at Document settings already, but this seems to do nothing to the file upload. There is still the focus retrieval process active. And it seems to just disable reranking, what we basically require for Knowledge collection retrieval. I would love to have everything towards Knowledge-Collections as it is currently with reranking as this works best. But for single Document uploading (small documents with just some pages) the retrieval process is working worse then full document mode. So a possibility to switch that behavior to make full document mode default (by env. variable) would be a great addition for the project. The token limit option that i mentioned in the Alternatives Considered section on top would be the cherry on top. Thx, best regards, Robert

GiteaMirror commented

2025-11-11 17:05:09 -06:00

@Classic298 commented on GitHub (Oct 24, 2025):

Did you try "enable full context mode" too in the admin document settings?

@Classic298 commented on GitHub (Oct 24, 2025): Did you try "enable full context mode" too in the admin document settings?

GiteaMirror commented

2025-11-11 17:05:09 -06:00

@deliciousbob commented on GitHub (Oct 24, 2025):

Yes I did test that just that moment again (v.6.34)
Activated that in Settings -> Documents -> "enable full context mode"

then uploaded a document and checked to mode by clicking onto the file:

It does the "get 3 questions" still, but actually it does pass the full context to the LLM.
Just checked the LiteLLM Logs. Showed me 65K tokens on the request.

But for Knowledge Collection retrieval this is causing retrieval of ALL Documents from the Collection:

So the "enable full context mode" does actually work for Document upload (only shows wrong if you click onto the file)
But i cannot activate it, as it would cause RAG for Knowledge Collections to not work anymore.

So, having an option to set that individually for file uploads and knowledge collections would be the desired feature.

@deliciousbob commented on GitHub (Oct 24, 2025): Yes I did test that just that moment again (v.6.34) Activated that in Settings -> Documents -> "enable full context mode" <img width="950" height="78" alt="Image" src="https://github.com/user-attachments/assets/2660e79e-6b7e-41a8-9bbe-8bd365ced155" /> then uploaded a document and checked to mode by clicking onto the file: <img width="150" height="70" alt="Image" src="https://github.com/user-attachments/assets/33ff190d-eef0-45fb-9dcb-a257a2df2c5d" /> It does the "get 3 questions" still, but actually it does pass the full context to the LLM. Just checked the LiteLLM Logs. Showed me 65K tokens on the request. But for Knowledge Collection retrieval this is causing retrieval of **ALL Documents** from the Collection: <img width="800" height="286" alt="Image" src="https://github.com/user-attachments/assets/5afefa95-6653-4439-8ba6-818cd8f48a49" /> So the **"enable full context mode" does actually work for Document upload** (only shows wrong if you click onto the file) But i cannot activate it, as it would **cause RAG for Knowledge Collections to not work anymore**. So, having an **option** to set that **individually** for **file** **uploads** and **knowledge collections** would be the desired feature.

GiteaMirror commented

2025-11-11 17:05:10 -06:00

@Classic298 commented on GitHub (Oct 24, 2025):

Understand. Do you think we should reopen the issue or for you to open a new feature request clearly explaining the feat you want? I think it would be better to clearly have it defined why full context mode in admin settings is not a solution

@Classic298 commented on GitHub (Oct 24, 2025): Understand. Do you think we should reopen the issue or for you to open a new feature request clearly explaining the feat you want? I think it would be better to clearly have it defined why full context mode in admin settings is not a solution

GiteaMirror commented

2025-11-11 17:05:10 -06:00

@deliciousbob commented on GitHub (Oct 24, 2025):

I can create a new Issue, that is fine for me too, describing the situation. :) thx

@deliciousbob commented on GitHub (Oct 24, 2025): I can create a new Issue, that is fine for me too, describing the situation. :) thx

GiteaMirror referenced this issue

2026-04-19 20:49:18 -05:00

[GH-ISSUE #6754] Error: (sqlite3.DataError) string or blob too big and sqlite3.OperationalError after updating from 0.3.28 to 0.3.35 #14474

GiteaMirror referenced this issue

2026-04-25 04:20:23 -05:00

[GH-ISSUE #6754] Error: (sqlite3.DataError) string or blob too big and sqlite3.OperationalError after updating from 0.3.28 to 0.3.35 #30002

GiteaMirror referenced this issue