Mintplex-Labs/anything-llm
1
0
Fork 0
mirror of https://github.com/Mintplex-Labs/anything-llm.git synced 2026-07-19 14:13:35 -04:00
Code Issues 328 Packages Projects Releases 10 Wiki Activity

[GH-ISSUE #4064] [BUG]: Bad Design Choices #2590

New Issue
Closed
opened 2026-02-22 18:30:22 -05:00 by yindo · 2 comments
yindo commented 2026-02-22 18:30:22 -05:00
Owner
Copy Link
Copy Source

Originally created by @digitalassassins on GitHub (Jun 27, 2025).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/4064

How are you running AnythingLLM?

All versions

What happened?

If you set your workspace to "Query" mode, it will not return a response unless there is a Document Context

If you then upload a PDF document to the chat window and ask a question on the document, even though the mode is switched to Query, and you have explicitly provided a document and tried to ask a question on the document you have just uploaded. It won't work..

Why?

Because having the Workspace mode on Query AND providing a document to AnythingLLM isn't enough.

You also must put @agent before the message AND provide the filename exactly as it is listed (case sensitive).

https://docs.anythingllm.com/llm-not-using-my-docs#llms-do-not-introspect

If you don't provide the filename exactly, it queries the whole workspace..

So you have to:

  • Switch the workspace to "Query" for the Documents context only
  • Upload a file
  • Provide the exact file name of the document you have just uploaded 🙃
  • AND put @agent before every message..

The most backwards and user-unfriendly system I've used.

Should just go the whole hog and make it, plus the user must stand on one leg and touch their nose with their left hand while clicking the mouse button..

Who comes up with these design ideas? 🫤

Are there known steps to reproduce?

No response

Originally created by @digitalassassins on GitHub (Jun 27, 2025). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/4064 ### How are you running AnythingLLM? All versions ### What happened? If you set your workspace to "Query" mode, it will not return a response unless there is a Document Context If you then upload a PDF document to the chat window and ask a question on the document, even though the mode is switched to Query, and you have explicitly provided a document and tried to ask a question on the document you have just uploaded. It won't work.. Why? Because having the Workspace mode on Query AND providing a document to AnythingLLM isn't enough. You also must put @agent before the message AND provide the filename exactly as it is listed (case sensitive). https://docs.anythingllm.com/llm-not-using-my-docs#llms-do-not-introspect If you don't provide the filename exactly, it queries the whole workspace.. So you have to: + Switch the workspace to "Query" for the Documents context only + Upload a file + Provide the exact file name of the document you have just uploaded 🙃 + AND put @agent before every message.. The most backwards and user-unfriendly system I've used. Should just go the whole hog and make it, plus the user must stand on one leg and touch their nose with their left hand while clicking the mouse button.. Who comes up with these design ideas? 🫤 ### Are there known steps to reproduce? _No response_
yindo added the possible bug label 2026-02-22 18:30:22 -05:00
yindo closed this issue 2026-02-22 18:30:22 -05:00
yindo commented 2026-02-22 18:30:22 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Jun 27, 2025):

Not the case, this is just the difference between RAG chat and agentic chat. You are asking meta-questions, most likely instead of asking questions about the content (which is what RAG is). We have a whole doc about this, but nobody seems to search their GH issue before posting:

https://docs.anythingllm.com/llm-not-using-my-docs

Additionally, https://github.com/Mintplex-Labs/anything-llm/pull/3483 would address this as we can then start to move towards always agent mode. The reason that is not the case is because people expect a 0.6B model to do full text comprehension on a 4000-page PDF and are surprised when it doesn't work. Sure, there is work to do to improve that UX and we are doing that as time goes on.

In the meantime, you can fix it on a fork if our design decisions are unbearable to you, since you seem to have a perfect idea of how it should work so hopefully you can contribute back once its done - looking forward to the PR!

@timothycarambat commented on GitHub (Jun 27, 2025): Not the case, this is just the difference between RAG chat and agentic chat. You are asking meta-questions, most likely instead of asking questions about the content (which is what RAG is). We have a whole doc about this, but nobody seems to search their GH issue before posting: https://docs.anythingllm.com/llm-not-using-my-docs Additionally, https://github.com/Mintplex-Labs/anything-llm/pull/3483 would address this as we can then start to move towards _always agent_ mode. The reason that is not the case is because people expect a 0.6B model to do full text comprehension on a 4000-page PDF and are surprised when it doesn't work. Sure, there is work to do to improve that UX and we are doing that as time goes on. In the meantime, you can fix it on a fork if our design decisions are unbearable to you, since you seem to have a perfect idea of how it should work so hopefully you can contribute back once its done - looking forward to the PR!
yindo commented 2026-02-22 18:30:22 -05:00
Author
Owner
Copy Link
Copy Source

@digitalassassins commented on GitHub (Jun 27, 2025):

I already posted that link in the message, and I did read it.

Discord has people complaining it isn't working in 1.8.2

OpenWebUI and GPT4All don't require you to:

  • Select Query Mode
  • Upload a Document
  • Copy the exact Filename for the document you have just uploaded into the chat window
  • and type @agent before the message.

The design choices are backwards, surely if the mode is in "Query mode" you know the user is wanting to query their documents, if they have provided a upload within the chat window file upload box.. then correct me if I'm wrong, but they probably want to chat with the document they just uploaded, not the whole workspace.

Why put 4 extra steps into the process and then fail if you don't meet all of the criteria?

If they wanted the chat with the document as part of the workspace, they would just upload the document to the workspace and not the chat window?

Surely you can ascertain from the fact that the Workspace is in Query mode and there is a document that has just been added to the chat window, that this user wants to chat with the specific document they are uploading.

So you could then auto-add the filename and the @agent to the prompt in the background to make it user-friendly.. 🙃

Like I said, Bad design decisions..

When you have to explain it over and over again to people who can't work it out. You will eventually change it so it is more user-friendly..

And your outlook is that you are not wrong because you have made it complicated for the user to use.
No, the user is wrong because they haven't read lots of guides posted on the website and deciphered the puzzle yet.

The fact that you have to write a user guide explaining why the user is wrong in the way they are using the RAG system says it all..

@digitalassassins commented on GitHub (Jun 27, 2025): I already posted that link in the message, and I did read it. Discord has people complaining it isn't working in 1.8.2 OpenWebUI and GPT4All don't require you to: - Select Query Mode - Upload a Document - Copy the exact Filename for the document you have just uploaded into the chat window - and type @agent before the message. The design choices are backwards, surely if the mode is in "Query mode" you know the user is wanting to query their documents, if they have provided a upload within the chat window file upload box.. then correct me if I'm wrong, but they probably want to chat with the document they just uploaded, not the whole workspace. Why put 4 extra steps into the process and then fail if you don't meet all of the criteria? If they wanted the chat with the document as part of the workspace, they would just upload the document to the workspace and not the chat window? Surely you can ascertain from the fact that the Workspace is in Query mode and there is a document that has just been added to the chat window, that this user wants to chat with the specific document they are uploading. So you could then auto-add the filename and the @agent to the prompt in the background to make it user-friendly.. 🙃 Like I said, Bad design decisions.. When you have to explain it over and over again to people who can't work it out. You will eventually change it so it is more user-friendly.. And your outlook is that you are not wrong because you have made it complicated for the user to use. No, the user is wrong because they haven't read lots of guides posted on the website and deciphered the puzzle yet. The fact that you have to write a user guide explaining why the user is wrong in the way they are using the RAG system says it all..
yindo changed title from [BUG]: Bad Design Choices to [GH-ISSUE #4064] [BUG]: Bad Design Choices 2026-06-05 14:47:26 -04:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
Desktop
Docker
Integration Request
Integration Request
OS: Linux
OS: Mobile
OS: Windows
UI/UX
blocked
bug
bug
core-team-only
documentation
duplicate
embed-widget
enhancement
feature request
github_actions
good first issue
investigating
needs info / can't replicate
possible bug
pull-request
Mirrored from GitHub Pull Request
 question
stage: specifications
wontfix
No Label possible bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#2590
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?