mirror of
https://github.com/open-webui/open-webui.git
synced 2026-07-16 23:21:44 -05:00
[GH-ISSUE #24851] issue: OWUI parrots the user in roleplay scenarios #123721
Reference in New Issue
Block a user
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?
Originally created by @frenzybiscuit on GitHub (May 17, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/24851
Check Existing Issues
Installation Method
Docker
Open WebUI Version
0.9.5
Ollama Version (if applicable)
N/A
Operating System
Linux
Browser (if applicable)
Firefox
Confirmation
README.md.Expected Behavior
The LLM should follow the system prompt and not parrot/speak for user.
Actual Behavior
It parrots the user.
Steps to Reproduce
Use the following narrator prompt:
https://gist.github.com/frenzybiscuit/23050d9cffbc67962bad40239089994f
Use the following parameters:
Logs & Screenshots
OWUI:
Sillytavern with the exact same prompt (and chat completion!):
Additional Information
N/A
@owui-terminator[bot] commented on GitHub (May 17, 2026):
🔍 Related Issues Found
I found some existing issues that might be related. Please check if any of these are duplicates or contain helpful solutions:
🟣 #17198 issue: Duplicated system prompt in the request
This is directly about system prompt handling in the request path. The bug there shows OWUI duplicating/misplacing system instructions, which is the same family of prompt-routing issues causing roleplay behavior to break here.
by DanielusG ·
bug🟣 #19049 feat: general system prompt shouldn't overwrite system prompt of custom model
This issue reports that a general system prompt overrides a custom model's more specific system prompt. If the roleplay narrator prompt is being ignored or flattened, this is a likely underlying cause in how system prompts are selected and applied.
by burntcouscous
🟣 #22965 issue: Code interpreter prompt (Pyodide) injected into user turn instead of system prompt
This closed bug is about prompt text being appended to the user turn instead of the system prompt, which is very similar to the current complaint that the model treats narrator/system instructions as user content and parrots them.
by AirRunner ·
bug💡 If your issue is a duplicate, please close it and add any additional details to the existing issue instead.
This comment was generated automatically. React with 👍 if helpful, 👎 if not.
@Classic298 commented on GitHub (May 17, 2026):
there's ... what?
Open WebUI doesn't do anything. It just sends whatever you give it to the AI. Maybe sillytavern has some additional prompts? did you check? did you intercept the raw requests being sent by Open WebUI and compared them to the raw request of whatever ST sends?
@frenzybiscuit commented on GitHub (May 17, 2026):
Yes, I do see a difference.
In SillyTaverns replies, it doesn't send the thinking content back.
In OpenWebUI, it does.
Infact, it's odd. Because I have strip thinking filters enabled..
SillyTavern:
OWUI: