Paste Text as Attachment-Like Elements

[mawburn]

Feature Description

This feature would treat large blocks of pasted text as attachment-like elements in the chat interface. When a user pastes a substantial amount of text (code, logs, or other text-based content), instead of displaying the entire content directly in the chat window, the system would represent it as an attachment-style element similar to uploaded files.

Key Technical Note

While the interface resembles file attachments, this feature would not upload the pasted content as a file. Instead, the text is appended directly to the prompt, maintaining a cleaner UI without changing how the text is processed by the model.

Implementation Details

This feature would significantly improve the user experience when working with large text blocks. Currently, pasting large content makes the prompt difficult to work with as it fills up the chat window, especially when dealing with multiple blocks of text.

The implementation would:

• Automatically detect when pasted content exceeds a certain size threshold (the exact threshold is open for discussion)
• Convert large pastes into an attachment-like element in the chat interface
• Allow users to click on the element to view the full content in a modal
• Process the pasted content as part of the prompt text, not as a file upload
• Include a user setting to toggle this feature on/off

Visual Design

This first mockup shows what the feature would look like in the normal chat interface. Large pasted text chunks are treated like other uploaded attachments, with a distinct visual indicator showing it's "Pasted from clipboard" rather than an uploaded file. This keeps the chat interface clean and organized.

The second mockup demonstrates what happens when a user clicks on the pasted content element. A modal opens displaying the full text content, making it easy to review the pasted material before sending.

Screen.Recording.2025-04-29.at.2.22.13.PM.mov

A short video demonstration showing the interaction flow when clicking on the pasted content element in the chat interface.

This approach maintains a clean chat interface while still allowing users to work with large blocks of text efficiently. The feature is inspired by similar functionality in other LLM providers but designed specifically to fit LibreChat's existing UI/UX patterns.

Implementation Readiness

We already have developers at our company ready to build this feature and are currently working on prototypes for implementation.

cc: @nhtruong @sbruel @katbuilt

[danny-avila]

Thanks! Seems closely related: #7135

[mawburn]

@danny-avila I think that might be related? I'm not sure about the OCR part. It kind of sounds like if you grab a .txt file and paste it, then it doesn't do an upload, but I think that should stay an upload. Often when people in my org upload files, the files are too big for the context window and need to be part of RAG.

With how many models that can possibly be added and how often they change, configuring the context size to determine what should go into the RAG pipeline seems unnecessarily complicated.

This is really just about cleaning up the UI more than any functional changes to the way attachments or uploads work. The pasted text would just be part of the prompt itself. If the pasted text is too big for context, it should probably just error as usual.

....with an individual user toggle, of course. Not everyone would want a change like this, but it's very nice to have for those who do and would use it frequently.

[nhtruong]

Is there a reason why we don't want to treat this as a file upload? We can upload these blobs of text as .txt or .md files as is (i.e. without going through RAG) and let LibreChat append the text at the end of the prompt before send it to the AI endpoints. The logic that handles text files is already on the server.

[nhtruong]

There are 2 routes to handle this:

• Treat the pasted texts as extra fields on MongoDB
• Treat the pasted texts as uploaded txt files.

With MongoDB approach, the users doesn't have to involve a 3rd party storage solution to store the text files. With Uploaded File approach, maybe we can store these text files like we store images (by default: on Libre backend server).

It gets complicated for the MongoDB approach when we have to display these pasted texts alongside other uploaded files. We must:

• Preserve the order of the attachments even though we now have 2 different classes of attachments (from uploaded files, and from MongoDB)
• Add logic to display, preview, and delete the pasted text attachments (This comes for free if they are treated as files like other attachments)

With the text file approach we have a couple obstacles:

• Cluttering the Attache Files list. Though I think we can hide pasted text files by default.
• Uploading plain-text as-is is current not yet supported. This is not a big lift from what I've gathered. We just need to add a path to upload without a tool_resource.

What's your take on this, @danny-avila?