[GH-ISSUE #2056] [BUG]: Context not correct after embedding file #1339

Closed
opened 2026-02-22 18:24:20 -05:00 by yindo · 1 comment
Owner

Originally created by @florian-hansen-mw on GitHub (Aug 6, 2024).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/2056

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

Hi,

I am on the AnythingLLM desktop app for MacOS with Apple Silicon chips, v1.6.0
I am integrating it with a local Ollama installation running the mistral-nemo:latest model with a max-tokens config of 120.000.
I use the default Vector DB and Embedder and Text Splitter.

I have configured a workspace with a Prompt "You are alf". I have changed nothing else. Regarding the question "Who are you", the model correctly replies with "I am Alf..."
image

Now I create a new Thread in the same Workspace.
I added a file (content should not matter, it's a 6 line shell script).
When I now ask "Who are you?" it still knows it's Alf.
Now I appended a 6ki file (second message), and the model forgets, who it is:

image

Are there known steps to reproduce?

No response

Originally created by @florian-hansen-mw on GitHub (Aug 6, 2024). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/2056 ### How are you running AnythingLLM? AnythingLLM desktop app ### What happened? Hi, I am on the AnythingLLM desktop app for MacOS with Apple Silicon chips, v1.6.0 I am integrating it with a local Ollama installation running the mistral-nemo:latest model with a max-tokens config of 120.000. I use the default Vector DB and Embedder and Text Splitter. I have configured a workspace with a Prompt "You are alf". I have changed nothing else. Regarding the question "Who are you", the model correctly replies with "I am Alf..." ![image](https://github.com/user-attachments/assets/1f99bda4-33a3-4400-b08b-40787c8189a1) Now I create a new Thread in the same Workspace. I added a file (content should not matter, it's a 6 line shell script). When I now ask "Who are you?" it still knows it's Alf. Now I appended a 6ki file (second message), and the model forgets, who it is: ![image](https://github.com/user-attachments/assets/019066af-b36b-4920-9aad-0c9d53742d18) ### Are there known steps to reproduce? _No response_
yindo added the possible bug label 2026-02-22 18:24:20 -05:00
yindo closed this issue 2026-02-22 18:24:20 -05:00
Author
Owner

@timothycarambat commented on GitHub (Aug 6, 2024):

This is an LLM behavior - this is not related to AnythingLLM or how it functions. Depending on the quantization of your model it may not even "listen" to your system prompt at all.

@timothycarambat commented on GitHub (Aug 6, 2024): This is an LLM behavior - this is not related to AnythingLLM or how it functions. Depending on the quantization of your model it may not even "listen" to your system prompt at all.
yindo changed title from [BUG]: Context not correct after embedding file to [GH-ISSUE #2056] [BUG]: Context not correct after embedding file 2026-06-05 14:40:14 -04:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#1339