RAG on uploaded files / web search can only work for the first chat #1837

New Issue

GiteaMirror · 2025-11-11T14:54:28-06:00

GiteaMirror commented

2025-11-11 14:54:28 -06:00

Originally created by @foxworld306 on GitHub (Aug 21, 2024).

Bug Report

Installation Method

[Describe the method you used to install the project, e.g., git clone, Docker, pip, etc.]

Environment

**Open WebUI Version: 0.3.13
**Ollama (if applicable): 0.3.5
** Web search tool: Searxng:latest
**Operating System: Ubuntu 22.04
**Browser (if applicable): Chrome, Edge

Confirmation:

[x ] I have read and followed all the instructions provided in the README.md.
[x ] I am on the latest version of both Open WebUI and Ollama.
I have included the browser console logs.
[ x] I have included the Docker container logs.
[x ] I have provided the exact steps to reproduce the bug in the "Steps to Reproduce" section below.

Expected Behavior:

Continuous chat on file content
Using web search multiple times in a conversation session

Actual Behavior:

I found that in the Open Webui environment I deployed, it is not possible to continuously have RAG conversations with uploaded documents. It only works for the first conversation in the New Chat session I opened. From the second conversation onwards, it seems that I have never uploaded any documents. But when I look at the logs of Open Webui, I can see that the backend has already provided document retrieval results. It seems that starting from the second conversation, the retrieval results were not provided to LLM as expected, only LLM's knowledge was used to answer.
This issue also affects my online search operations. I must perform the search action in the first conversation of each New Chat session, otherwise starting from the second conversation, even if I can see Searxing returning the search results, I won't be able to get the correct answer.

I have already tried:

Place the document in workspace ->documents
Using different LLM models (Qwen2 and GLM4)
reset vector storage
Using or not using Hybrid search,
Different Embedding Models
Different Rerank models
Using different web search methods, including searxng, google pse, duckduckgo, tavily.
Deploy different Open Webui (main / cuda)

All the above operations cannot solve this problem.

Can you please help to check?

Logs and Screenshots

Additional Information

[Include any additional details that may help in understanding and reproducing the issue. This could include specific configurations, error messages, or anything else relevant to the bug.]

Note

If the bug report is incomplete or does not follow the provided instructions, it may not be addressed. Please ensure that you have followed the steps outlined in the README.md and troubleshooting.md documents, and provide all necessary information for us to reproduce and address the issue. Thank you!

Originally created by @foxworld306 on GitHub (Aug 21, 2024). # Bug Report ## Installation Method [Describe the method you used to install the project, e.g., git clone, Docker, pip, etc.] ## Environment - **Open WebUI Version: 0.3.13 - **Ollama (if applicable): 0.3.5 - ** Web search tool: Searxng:latest - **Operating System: Ubuntu 22.04 - **Browser (if applicable): Chrome, Edge **Confirmation:** - [x ] I have read and followed all the instructions provided in the README.md. - [x ] I am on the latest version of both Open WebUI and Ollama. - [ ] I have included the browser console logs. - [ x] I have included the Docker container logs. - [x ] I have provided the exact steps to reproduce the bug in the "Steps to Reproduce" section below. ## Expected Behavior: - Continuous chat on file content - Using web search multiple times in a conversation session ## Actual Behavior: I found that in the Open Webui environment I deployed, it is not possible to continuously have RAG conversations with uploaded documents. It only works for the first conversation in the New Chat session I opened. From the second conversation onwards, it seems that I have never uploaded any documents. But when I look at the logs of Open Webui, I can see that the backend has already provided document retrieval results. It seems that starting from the second conversation, the retrieval results were not provided to LLM as expected, only LLM's knowledge was used to answer. This issue also affects my online search operations. I must perform the search action in the first conversation of each New Chat session, otherwise starting from the second conversation, even if I can see Searxing returning the search results, I won't be able to get the correct answer. I have already tried: - Place the document in workspace ->documents - Using different LLM models (Qwen2 and GLM4) - reset vector storage - Using or not using Hybrid search, - Different Embedding Models - Different Rerank models - Using different web search methods, including searxng, google pse, duckduckgo, tavily. - Deploy different Open Webui (main / cuda) All the above operations cannot solve this problem. Can you please help to check? ## Logs and Screenshots ![image](https://github.com/user-attachments/assets/64128dcd-5e25-4a49-889c-299d412bbc64) ![image](https://github.com/user-attachments/assets/3b928152-c107-4aa2-a6ac-2532d3838e2b) ![image](https://github.com/user-attachments/assets/81fffbac-af4d-4047-ba2e-e0e9b2b9ccb2) ## Additional Information [Include any additional details that may help in understanding and reproducing the issue. This could include specific configurations, error messages, or anything else relevant to the bug.] ## Note If the bug report is incomplete or does not follow the provided instructions, it may not be addressed. Please ensure that you have followed the steps outlined in the README.md and troubleshooting.md documents, and provide all necessary information for us to reproduce and address the issue. Thank you!

GiteaMirror closed this issue

2025-11-11 14:54:29 -06:00

GiteaMirror referenced this issue

2025-11-11 17:31:25 -06:00

[PR #1843] [CLOSED] Increase-Context-Length #7608

GiteaMirror referenced this issue

2025-11-11 17:31:44 -06:00

[PR #1881] [MERGED] 0.1.123 #7621

Sign in to join this conversation.