Conversation segmentation uses confusing UI and breaks context continuity [v0.7.7+]

[Kadesy]

🔀 Feature or Bug? Conversation segmentation uses confusing UI and breaks context continuity [v0.7.7+]

Problem Summary

Since v0.7.7, LibreChat automatically segments conversations under certain conditions, but uses identical arrow navigation as message alternatives. This creates significant UX confusion where users believe their conversation history has disappeared.

User Experience Impact

I've encountered this behavior since v0.7.7 and initially thought the platform was losing my data, which significantly impacted my confidence in using LibreChat. While LibreChat is otherwise exceptional, this UX issue creates unnecessary friction and user anxiety.

Current Behavior

• Arrow navigation (← 1/2 →) appears at the top of conversations for segment navigation
• Previous conversation segments become hidden by default
• Context is not carried over between segments
• Same UI pattern used for both message alternatives and conversation segments

Expected Behavior

• Clear distinction between message alternatives and conversation segments
• Visible indication when conversations are segmented and why
• Context preservation between segments, or clear communication about context loss
• Intuitive navigation that doesn't cause confusion

Possible Triggers

• Extended conversations (context window limits)
• Changing models mid-conversation
• Specific model configurations
• Time-based factors

Environment

• Versions: v0.7.7+ (currently v0.7.9-rc1)
• Platform: Remote Docker Compose deployment
• Access: Custom domain via Firefox/Brave on Android OnePlus 12

Clarification Needed

Is this segmentation behavior an intentional feature for managing:

• Context window limitations
• Model switching scenarios
• Performance optimization

If so, could we get clarification on:

• When segmentation triggers occur
• Whether context preservation is technically possible
• If user configuration options exist

Suggested Improvements

1. Distinct UI patterns for conversation segments vs message alternatives
2. User notification when segmentation occurs with brief explanation
3. Context preservation between segments where technically feasible
4. Documentation explaining this behavior in user guides
5. Visual indicators showing segment boundaries or reasons for segmentation

Additional Context

This appears to serve a valid technical purpose, but the current implementation can create user confusion and concern about data loss. Improving the UX around this functionality would greatly enhance user confidence and platform usability.

Version Information

docker images | grep librechat
ghcr.io/danny-avila/librechat-dev-api
latest cd3432a5c3f6 43 hours ago
1.18GB
ghcr.io/danny-avila/librechat-rag-api-dev-lite
latest 8858d2bd9ff5 10 days ago
1.48GB
ghcr.io/danny-avila/librechat-api
latest d7e7a503578e 2 weeks ago
1.19GB

Steps to Reproduce

Reproduction Steps (Current Understanding)

Confirmed Triggers

1. Start a long conversation in LibreChat
2. Change models during the extended conversation (high probability trigger)
3. Continue conversing until context limits may be reached

How to Discover the Issue

1. Notice conversation appears to suddenly "reset" or lose previous history
2. Scroll to the very top of the chat window
3. Look for numerical arrow navigation (e.g., ← 1/2 → or ← 2/2 →)
4. Click the arrows to navigate between hidden conversation segments
5. Observe that context from previous segments is not available in current segment

Uncertainty Note

⚠️ Exact reproduction conditions unclear - may also involve:

• Specific context token thresholds
• Particular model combinations
• Session duration factors
• Browser/platform specific behaviors

Versions Affected

• Confirmed since v0.7.7
• Currently observed in v0.7.9-rc1

Key Point: The segmentation appears to happen automatically without user notification, and the navigation UI is identical to message alternative arrows, causing the confusion.

What browsers are you seeing the problem on?

Firefox, Chrome

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Thanks for the report. While I appreciate the details, the immediate issue is not clear (reality vs. expectation). It would help me understand better if you could export an example conversation (top-right share button, export option, and use recursive JSON with all options default/checked), as well as provide more details on the exact models you are using.

Change models during the extended conversation (high probability trigger)

Different models have different context windows. Context from the earlier part of a large conversation are automatically discarded if the context window starts being approached/exceeded for current requests. However, this has no impact on the UI, as it all happens server-side.

As far as the context goes, what you see in the UI is what will be attempted to fit in the current context window, which will vary with each model selected.

numerical arrow navigation (e.g., ← 1/2 → or ← 2/2 →)

I call these "segments" branches, as I understand them to be branches of the main thread. They should only be created when using regeneration, or resubmitting an edited or previously sent message. If you are observing "branching" outside of these expected "triggers", you may be observing an edge case in the conversation flow. I personally switch models a lot even in long conversations, and have not observed this behavior outside of the expected triggers.

[Kadesy]

Hey! Thanks for the reply 🙏 apologies for the wrong terminology... I understand forking and branches, but what I'm experiencing doesn't seem like normal behaviour as you've defined it.

I’ve got a long chat JSON and a short video showing the issue clearly, but I’d prefer to share them privately since they include some sensitive info.

Is there a secure way I can send them to you directly? We can continue the conversation based on the findings here after.

[danny-avila]

You can DM me on discord:
https://discord.com/invite/librechat-1086345563026489514

[Kadesy]

Done 🤘🏽