feat: Enable Passing Uploaded Files Directly to Models #5763

New Issue

GiteaMirror · 2025-11-11T16:32:58-06:00

GiteaMirror commented

2025-11-11 16:32:58 -06:00

Originally created by @lajdzsi on GitHub (Jul 15, 2025).

Check Existing Issues

I have searched the existing issues and discussions.

Problem Description

When uploading a file to a chat, even when the "Bypass Embedding and Retrieval" option is turned on, the file's content is extracted and only the extracted text is passed to the model fully.
There are cases when it's beneficial to pass the files directly to the LLMs, and it is a standard feature in many similar solutions.

Desired Solution you'd like

I’d like to suggest adding an option to pass uploaded files directly to the models, bypassing automatic text extraction.
This could be e.g. a switch next to the file upload UI element.
The binary content of the files is already stored in the database, and can be accessed via the body dictionary, so the data is there.

Alternatives Considered

No response

Additional Context

Passing raw files would enable more flexible and powerful interactions, as well as the support of many different file formats. Currently, OpenWebUI only supports the passing of extracted text from uploads, which limits working with binary files, images, or structured data.

Implementing this feature would give users greater control and unlock new use cases.
Thanks for considering this enhancement!

Originally created by @lajdzsi on GitHub (Jul 15, 2025). ### Check Existing Issues - [x] I have searched the existing issues and discussions. ### Problem Description When uploading a file to a chat, even when the "Bypass Embedding and Retrieval" option is turned on, the file's content is extracted and only the extracted text is passed to the model fully. There are cases when it's beneficial to pass the files directly to the LLMs, and it is a standard feature in many similar solutions. ### Desired Solution you'd like I’d like to suggest adding an option to pass uploaded files directly to the models, bypassing automatic text extraction. This could be e.g. a switch next to the file upload UI element. The binary content of the files is already stored in the database, and can be accessed via the body dictionary, so the data is there. ### Alternatives Considered _No response_ ### Additional Context Passing raw files would enable more flexible and powerful interactions, as well as the support of many different file formats. Currently, OpenWebUI only supports the passing of extracted text from uploads, which limits working with binary files, images, or structured data. Implementing this feature would give users greater control and unlock new use cases. Thanks for considering this enhancement!

GiteaMirror closed this issue

2025-11-11 16:32:58 -06:00

GiteaMirror commented

2025-11-11 16:32:59 -06:00

@rgaricano commented on GitHub (Jul 15, 2025):

The way to pass raw files is to base64-encode them and send them as a response, which will result in a substantial increase in tokens and context length; this is not a viable solution.
Another option is to simply pass the upload URL to the model or share the document with the model via another URL that is accessible to it.

@rgaricano commented on GitHub (Jul 15, 2025): The way to pass raw files is to base64-encode them and send them as a response, which will result in a substantial increase in tokens and context length; this is not a viable solution. Another option is to simply pass the upload URL to the model or share the document with the model via another URL that is accessible to it.

GiteaMirror commented

2025-11-11 16:32:59 -06:00

@GmodCentral commented on GitHub (Jul 15, 2025):

Passing the URL makes most sense. I would really love this feature. Tools like python docx could make good use of this. Currently my solution is a bit convoluted as it involves pulling the latest filename from the API, then copying the file for editing into the /tmp dir. You can see how this works here.

https://gitlab.com/gmod-central/owui-word-doc-template-editor

It would make things a lot easier if the URL could be just fed directly into context when uploading a file. Like the file name and path.

@GmodCentral commented on GitHub (Jul 15, 2025): Passing the URL makes most sense. I would really love this feature. Tools like python docx could make good use of this. Currently my solution is a bit convoluted as it involves pulling the latest filename from the API, then copying the file for editing into the /tmp dir. You can see how this works here. https://gitlab.com/gmod-central/owui-word-doc-template-editor It would make things a lot easier if the URL could be just fed directly into context when uploading a file. Like the file name and path.

GiteaMirror referenced this issue

2025-11-11 18:12:41 -06:00

[PR #8580] [MERGED] Split the storage providers into separate classes in preparation for adding more cloud providers #9048

Sign in to join this conversation.

Branches Tags

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/open-webui#5763