QOL Suggestions From A Lettuce Fan

[sighlent07]

First, I want to say that I'm really enjoying Lettuce. I’ve pretty much started using it in place of ST because it feels lightweight and no fuss. I have also tried multiple alternative private frontends out there, and of all of them, this is the one I really gravitated to the most.

I really hope development continues, because what you have now is a very good foundation. Most of what I find myself missing now consists of quality-of-life improvements.

Here are some of my suggestions:

• Faster model switching directly from a generated message
  There is already a quick-change option for switching models, but it currently requires opening another panel and selecting the replacement model from there. It would be very convenient if the model info shown beneath the generated response could itself become a dropdown. For example, clicking deepseek-v4-pro could open the model list, and selecting another model could immediately regenerate that response using the newly selected model. This would make model comparisons and testing different responses substantially faster.
  Here's a mockup:

• Make the Chats page display actual conversations only
  At the moment, the Chats page appears to include characters even when no conversation has been opened with them. Ideally, it would be much easier to handle if the page could display only characters with and existing chat, have sorting options based on recent usage, etc.

• Sorting and tag filtering for the Library page
  Since the Library is meant to contain all characters, personas, lorebooks and etc., the full collection belongs there. However, it would benefit greatly from stronger organization controls, like sorting (alphabetical, time of addition/creation) and filtering (by tags, creator, etc.)

• A global default for Dynamic Memory
  I prefer using Dynamic Memory for nearly every chat, but because Manual Memory is currently the default, I have to enable Dynamic Memory separately whenever I set up a new character or conversation. Could there be a global setting for it? I think this could work similarly to assigning a default model while still allowing individual chats to override the global choice.

• User control over the context budget
  I may have missed an existing setting, so part of this is also a question: Does Lettuce currently impose a fixed or automatic limit on how much of a model’s available context window it uses? The documentation explains what context length is and how Dynamic Memory limits the amount of stored memory injected into the prompt, but I could not find a user-facing control for choosing the context budget itself. It would be useful to have an optional advanced setting that controls the maximum context sent to a model, either globally, per provider, per model, or per chat. For example, I may sometimes want to restrict DeepSeek to around 40k context for cost and efficiency, while at other times I may intentionally want to use 100k+ context for a very long roleplay or analysis session.

• More control over speaker selection in group chats
  In my experience with group chats so far, one character appears to be selected to reply during each turn. I would love an additional option that allows multiple members to participate within one generation, either chosen by LLM or set by the user.

• More visible alternate-greeting navigation
  Characters with alternate greetings do not appear to have an obvious next/previous control in the chat interface. I eventually discovered through trial and error that the greeting can be changed by clicking and dragging or swiping across the chat bubble. However, this interaction is not immediately discoverable when using a mouse. A more visible indication to navigate through alt greetings would be very helpful.

• Use Creator Notes for imported character previews
  When importing characters, the displayed description currently seems to be populated using a truncated portion of the character definition. That is not always an ideal preview, since the definition is often written as a prompt instead of a description. It can also cut off at an awkward point. For formats that include Creator Notes, it may be a better source for the card description because creators commonly use that field for the character’s hook, premise, content info, or intro summary.

---

None of these issues has stopped me from enjoying the app. I am just suggesting these because I think it would improve the overall experience in using Lettuce based on my own standpoint. I am very thrilled to have found this project. Thank you for building this, hopefully this gives you an insight on how users like me use your app.

Thank you for reading this and hearing me out!

[MegalithOfficial]

Thanks for taking the time to write such detailed feedback! Here are my thoughts on each suggestion:

1. The Chats page intentionally includes characters without an existing session. Opening one automatically creates the first session. If you do not want a character displayed there, you can use Hide Character. The Library remains the complete collection.
2. Faster model switching sounds useful, but I want to keep the chat column clean. We could add it to the message action menu and offer it as a configurable widget. Making the model name beneath a response clickable could also provide a subtle shortcut without adding more visible controls.
3. Could you elaborate on how you would like Library sorting and filtering to work? For example, which sorting options and tag or creator filtering workflow would be most useful to you?
4. A global default for Dynamic Memory makes sense. We can still retain per-character and per-chat overrides.
5. The alternate-greeting navigation sounds like a bug or discoverability issue. I’ll investigate and fix it.
6. Context limiting is complicated because many providers do not support an explicit context-length parameter. Dynamic Memory exists partly to manage long conversations without sending the entire history every time, so a universal context-budget setting might not behave consistently across providers.
7. Multiple speakers per turn has been discussed in the community. At the moment, something similar can be directed manually through Director mode. Fully automatic multi-character turns would require more design work around speaker order and generation behavior.
8. Using Creator Notes as a better fallback for imported character previews makes sense. I’ll fix that.

Thanks again for the thoughtful suggestions and for supporting Lettuce! Please also join our Discord server if you haven’t already. We’d be happy to discuss these ideas further with you and the rest of the community.

[sighlent07]

Thanks for replying! :)

Sorting

With regards to the library sorting: I was thinking maybe just the basic sorting parameters: alphabetical, last added/recently created, in the future, if the app were to track how many distinct chats the character has, then maybe you could also sort it by number of chats.

Filtering

For the filtering, using the tag system would be the best way to implement this for me. There is already a field for tags after all, and maybe the user can set existing tags as filters (i.e. filtering on POV, Genre, Multiple Characters, etc.)

If filtering with unstructured tags is a mess, maybe you can limit it with default common tags first like 'Male', 'Female', 'AnyPOV', 'Smut', 'Fluff', etc. Usually there are tags common between platforms and you can just test it out with those first.

You could also go about this in a more free way where you can let the users set tags and set them as filters (like maybe have a 'manage tags and filters' panel)

Context Limiting

That makes sense, especially because not every provider exposes context length in the same way, and I understand that Dynamic Memory already helps manage long conversations without always sending the full chat history. But if it could be implemented as a provider-aware setting, even with labels like “only available when supported,” it would still be really appreciated.

The reason why I still think this would be a useful QOL feature despite the complexity is that, even with Dynamic Memory, some small details may not always be preserved in the way a user specifically needs. I’m the kind of person who uses a lot of callbacks and micro-continuity details in RP, so I often summarize tiny continuity signals in the narrative myself.

When I need to make or use that kind of summary, having access to more of a model’s full context capacity can really matter. Sometimes I want a chat to stay lightweight, but other times I would intentionally want to use a larger context window because the conversation depends on accumulated details.

Maybe it could work as something like:

• Auto / provider default
• Low context budget
• Medium context budget
• High context budget
• Custom token limit, if supported

Multiple Speakers

I'll check that work around out!

Discord

Thanks for the Discord invite, but I'm actually on a personal break from Discord right now. I'll probably join when I'm feeling okay to go back to Discord, but right now, I prefer discussing here. I hope that won't be a problem.

Thanks for hearing me out! 🥬

[MegalithOfficial]

Our discord server: https://discord.gg/Nkrc5rZZde