mirror of
https://github.com/open-webui/open-webui.git
synced 2026-07-16 14:39:31 -05:00
[GH-ISSUE #24522] issue: When continuing a conversation in the new version using a chat created in an older version, the system fails to send the full context to the model #107314
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 @CookSleep on GitHub (May 9, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/24522
Check Existing Issues
Installation Method
Docker
Open WebUI Version
v0.9.4
Ollama Version (if applicable)
No response
Operating System
Ubuntu 22.4
Browser (if applicable)
No response
Confirmation
README.md.Expected Behavior
When opening and continuing a chat history created in an older version using the latest release, the system sends the entire historical context to the model.
Actual Behavior
The system only sends a very limited amount of recent context to the model. The context window varies by conversation; some are restricted to the last three exchanges, while others can only include the single most recent turn.
Steps to Reproduce
After opening the latest version of OpenWebUI and selecting a chat history created in an older version, I tried to continue the conversation by asking the model about very specific and clear details mentioned earlier. The model completely forgot them, indicating that the chat history is not being sent in its entirety.
Logs & Screenshots
I'm very sorry, but I really couldn't find any error logs, and I'm also reluctant to post conversation content here that contains private information.
Additional Information
Maybe it's because the chat log structure was changed? I noticed this mentioned in the release notes.
@owui-terminator[bot] commented on GitHub (May 9, 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:
🟣 #22327 issue: Wrong chat history is sent to inference server
This bug reports that the wrong chat history is sent to the inference server, which is directly adjacent to the reported problem of incomplete context being sent after resuming a conversation. Both concern context assembly for requests.
by LostPhysx ·
bug🟢 #19594 feat: Persistent "Invisible" Metadata Object within Chat History for Context Management / OR new DB Table for storing these
This feature request is about storing persistent hidden metadata inside chat history for context management, which aligns with the user's suspicion that a chat log structure change may have altered how older chats preserve context.
by Classic298
🟢 #23781 issue: Recover full conversation history (reasoning + tools inputs/outputs) inside a tool
This issue is about missing prior tool-call history in later turns, another form of incomplete conversation context being provided to the model. It is related to context preservation, though specifically for tool execution.
by ccdv-ai ·
bug🟣 #20474 Chat infinitely loads when history.currentId points to orphaned/deleted message
This closed issue shows that malformed or inconsistent chat history state can break chat loading due to history.currentId/orphaned message problems. It is related insofar as older chats or migrated structures can cause conversation state issues, though it is not specifically about model context.
by MacJedi42 ·
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.
@CookSleep commented on GitHub (May 9, 2026):
I tested conversations created in the new version and didn't encounter the issues mentioned above. Therefore, I'm inclined to believe that the system might have some compatibility issues when reading conversation content from older versions.
@CookSleep commented on GitHub (May 9, 2026):
I haven't enabled any functions for managing context.
@CookSleep commented on GitHub (May 9, 2026):
You can also tell from the model response logs when continuing a conversation from an older version that the actual number of tokens sent doesn't match the full context of the conversation at all.
@CookSleep commented on GitHub (May 9, 2026):
I noticed that the old conversation with only one round of context was even more absurd—the very first message the model could see was actually its own last response, rather than my user message!
@tjbck commented on GitHub (May 10, 2026):
@CookSleep could you test the latest dev? it might already be addressed.
@2259819800 commented on GitHub (May 10, 2026):
I also encountered the same problem
@CookSleep commented on GitHub (May 10, 2026):
Thanks a lot! I tried the dev version, and everything is working perfectly now!