enh: direct full-document upload mode bypassing the embedding process #2382

New Issue

2025-11-11T15:06:03-06:00

GiteaMirror commented

2025-11-11 15:06:03 -06:00

Originally created by @g4challenge on GitHub (Oct 15, 2024).

Feature Request

The newly added full-text feature would allow users to upload a document and use it directly, without waiting for embedding.

Is your feature request related to a problem? Please describe.
It allows me for example, to directly and completely summarize a youtube video, without waiting for embeddings, and saves me time.

Describe the solution you'd like
Introduce a separate full-text upload option or a toggle, or use a double '##' to enable direct full-text usage.

Describe alternatives you've considered
none yet.

Originally created by @g4challenge on GitHub (Oct 15, 2024). # Feature Request The newly added full-text feature would allow users to upload a document and use it directly, without waiting for embedding. **Is your feature request related to a problem? Please describe.** It allows me for example, to directly and completely summarize a youtube video, without waiting for embeddings, and saves me time. **Describe the solution you'd like** Introduce a separate full-text upload option or a toggle, or use a double '##' to enable direct full-text usage. **Describe alternatives you've considered** none yet.

GiteaMirror closed this issue

2025-11-11 15:06:03 -06:00

GiteaMirror commented

2025-11-11 15:06:03 -06:00

@Naschkater commented on GitHub (Oct 23, 2024):

I think an option to create per model guidelines for bypassing RAG would be a great option too.

@Naschkater commented on GitHub (Oct 23, 2024): I think an option to create per model guidelines for bypassing RAG would be a great option too.

GiteaMirror commented

2025-11-11 15:06:04 -06:00

@Joni1717 commented on GitHub (Nov 6, 2024):

Or mabe an option to say only use the Rag If Minimum 3 documents uploaded.

@Joni1717 commented on GitHub (Nov 6, 2024): Or mabe an option to say only use the Rag If Minimum 3 documents uploaded.

GiteaMirror commented

2025-11-11 15:06:04 -06:00

@GrayXu commented on GitHub (Nov 25, 2024):

I think an option to create per model guidelines for bypassing RAG would be a great option too.

Strongly support this. Manually modifying it each time is very troublesome; a global switch to modify the default settings would be much better.

@GrayXu commented on GitHub (Nov 25, 2024): > I think an option to create per model guidelines for bypassing RAG would be a great option too. Strongly support this. Manually modifying it each time is very troublesome; a global switch to modify the default settings would be much better.

GiteaMirror commented

2025-11-11 15:06:04 -06:00

@BryceWG commented on GitHub (May 11, 2025):

Seven months later, I still want this feature. My core idea is to separate the embedding retrieval logic of the knowledge base from that of regular conversation file uploads, allowing independent configuration.

@BryceWG commented on GitHub (May 11, 2025): Seven months later, I still want this feature. My core idea is to separate the embedding retrieval logic of the knowledge base from that of regular conversation file uploads, allowing independent configuration.

GiteaMirror commented

2025-11-11 15:06:04 -06:00

@spammenotinoz commented on GitHub (May 13, 2025):

With these new models, "Bypass Embeddings and Retrieval" is frigging brilliant, the results are as good as ChatGPT.
I get the background into RAG and why some use cases, still require it, but heck try life without it the quality of the LLM responses is a massive improvement.

@spammenotinoz commented on GitHub (May 13, 2025): With these new models, "Bypass Embeddings and Retrieval" is frigging brilliant, the results are as good as ChatGPT. I get the background into RAG and why some use cases, still require it, but heck try life without it the quality of the LLM responses is a massive improvement. ![Image](https://github.com/user-attachments/assets/ff784473-bd0c-4765-b53d-38f88805dc2a)