Many people struggle to set up and use SillyTavern despite its comprehensive official documentation and various community-created guides there to assist them.
SillyTavern is a versatile frontend designed for power users. Its developers do not present it as a simple product tailored for any specific use case and instead encourage users to embrace its learning curve.
During RPWithAI’s conversation with Cohee, RossAscends, and Wolfsblvt, the developers of SillyTavern, we discussed the challenges of maintaining and developing such a versatile tool.
SillyTavern: A Versatile Frontend
People confuse SillyTavern as just a local frontend for AI roleplay. Using SillyTavern for AI roleplay is only one of the many ways to use this powerful tool.
Unlike other frontends that focus on a specific use case, such as a “local ChatGPT alternative” or for “immersive AI roleplay,” SillyTaven positions itself as a frontend that “allows you to interact with text generation LLMs, image generation engines, and TTS voice models.”
We don’t try to position ST as a “better alternative” to some other app, like “Character AI but uncensored” or “ChatGPT but local.”
Cohee, SillyTavern’s Developer, in a conversation with RPWithAI.
It doesn’t claim to be an easy alternative to other services or products. SillyTavern is a versatile, highly customizable frontend for power users, rewarding those who master its learning curve.
That’s one reason why people find it challenging to set up and use SillyTavern. They want to use it for a specific purpose, like AI roleplay, and expect it to work ‘out of the box’. Reading the official documentation can help new users solve most of the issues they face. However, many prefer instant solutions or a smoother setup experience, similar to what other newer frontends offer.
Docs are really important. 90% of “how do I” support cases could be solved by just throwing in a link to the docs.
Cohee, SillyTavern’s Developer, in a conversation with RPWithAI.
SillyTavern Is Powerful And Versatile, But It’s Just A Tool
SillyTavern is a powerful and versatile frontend, but it’s just a tool. A piece of the puzzle. It’s not an all-in-one solution. You still need a backend for LLM inference, a suitable model for your purpose, a basic understanding of how LLMs work, and more. SillyTavern can’t magically provide you with the best experience for AI roleplay or any other use case.
But that’s the thing about ST. It’s just the tool. While it is very powerful, most of the really important things happen outside of it. The API/backend, its functions. The prompts you provide. I don’t think any of that is inherently special to ST – besides the magic [of] how we build prompts from World Info and Prompt Manager, with macros, etc.
Wolfsblvt, SillyTavern’s Developer, in a conversation with RPWithAI.
That’s why SillyTavern positions itself as a frontend for power users. The developers expect you to have some basic knowledge and problem-solving skills, and don’t hold your hand through the process.
Users who are used to corporate models or tailored/limited AI RP chat experiences like Character AI want the next best thing to be “easy”, “high quality,” and “free” at the same time. In reality, you can really only expect two of the three. If something is easy and good, it’s going to cost you money. If you want easy and free, expect low quality. If you want free and high quality, expect to put in some work (this is ST’s lane).
RossAscends, SillyTavern’s Developer, in a conversation with RPWithAI.
If SillyTavern’s learning curve feels too steep, other frontends might be a simpler option for your specific use case. But they often lack the features that SillyTavern provides and may not be open-source or free.
Developing And Maintaining SillyTavern Is Also A Challenge
Due to its versatility, maintaining and developing SillyTavern can also be challenging. Contributors often suggest ideas to improve the frontend, and developers spend time helping them turn those ideas into reality. However, some ideas prove too complex to implement and end up being abandoned.
We had about five people who wanted to start rewriting ST UI (the coveted “SillyTavern 2”) from scratch, but they all mysteriously disappeared (not joking), probably after the realization of how much scope they have to cover.
Cohee, SillyTavern’s Developer, in a conversation with RPWithAI.
People often underestimate the amount of work required to implement some ideas or rewrite parts of the core functionality. Enthusiastic contributors bring ambitious plans to the table, but the scope and complexity usually prove too much to see them through.
Regularly, devs come in with huge ideas and tasks, asking to rewrite X or “why can’t we make this code better?” And while we always try to help and guide them, Cohee and I have seen enough people like this come (and go) to know that most of the time it’ll just fizzle out. Those arduous tasks nearly never work out.
Wolfsblvt, SillyTavern’s Developer, in a conversation with RPWithAI.
But that doesn’t mean SillyTavern isn’t improving. It’s just more challenging. The developers also focus on maintaining backward compatibility, which adds another layer of complexity when updating existing features.
While ST 2.0 is a meme, that doesn’t mean we don’t iterate and improve, though. Parts of ST get rewritten regularly, whenever there is time or a task touches it […] ST always had this huge focus on backwards compatibility.
Wolfsblvt, SillyTavern’s Developer, in a conversation with RPWithAI.
Users’ data will always be preserved or cleanly migrated. Functionality that existed and that people use, will stay – or at least be covered [to] the same extent with the replacement feature. Even on the code side, we try to keep functions compatible for any extension out there that may use them directly.
As you might have figured out, this makes any iteration and improvement of existing stuff much harder. And sometimes so hard that it’s just not worth it to touch it at all.
Other frontends that have specific use cases don’t face similar challenges. Developing a streamlined tool isn’t as complex, and modifying core functions is much easier. With SillyTavern, users may use a particular feature in hundreds of ways, whereas on other frontends, they might only use a feature in a few ways.
Simpler Alternatives And SillyTavern 2.0
SillyTavern has also inspired other free, open-source projects, with developers building frontends that promise a simpler, more streamlined alternative to SillyTavern. These projects provide users with a frontend that doesn’t have a steep learning curve, but they also lack many features and customization options.
Cohee shared his opinion, saying that any project launched today needs a significant head start to rival SillyTavern’s feature set or have a “killer feature” that instantly draws users. The hands-on experience of the project’s developer also plays a crucial role in its success.
You can vibe code a working prototype or app skeleton in a single evening, but it’s the tiny, seemingly insignificant details and the hands-on developer experience that make all the difference. I can’t count how many times we’ve had long Discord “battles” over trivial things like whether something should be separated by one or two newlines.
Cohee, SillyTavern’s Developer, in a conversation with RPWithAI.
Cohee also believes that in the current landscape, where LLMs are imperfect and change is constant, a simpler and streamlined frontend offering a curated experience can become stale quickly, which could hinder a project’s success.
In my opinion, a “curated experience” gets stale fast. With how obscenely imperfect modern LLMs still are, relying on a rigid, hardcoded workflow won’t get you very far. I’ve also seen a few projects experimenting with node-based workflows (similar to ComfyUI), which looks promising but still feels too niche to become the “next big thing.”
Cohee, SillyTavern’s Developer, in a conversation with RPWithAI.
Wolfsblvt stated that SillyTavern serves as a valuable example for developers to identify what the frontend does well, as well as its pitfalls that other projects should avoid. However, creating a new frontend to match SillyTavern’s features or rewriting it from scratch is a challenging task even for its current developers.
To reach a matching feature level, even in one of [the] many different strengths ST has, is gonna take a while for any dev. It would for us if we would rewrite ST. Just think about [the] diversity of supported APIs. Power of World Info. Character management. Power User settings/toggles. And so much more.
Wolfsblvt, SillyTavern’s Developer, in a conversation with RPWithAI.
RossAscends also chimed in, stating that a rewrite or “SillyTavern 2,” if it were ever to happen, could bring drastic changes for users and contributors.
I think the core paradigms and metaphors used would be different, so it might not even ‘feel’ like the same kind of app. It also would probably not be done in JS/node/plain file structure, which might make it harder for fresh contributors to jump in.
RossAscends, SillyTavern’s Developer, in a conversation with RPWithAI.
SillyTavern: A Versatile Frontend For Power Users
SillyTavern is a versatile frontend designed for power users. The developers offer a powerful sandbox tool that you can customize as you like, and they are unapologetic about its steep learning curve. SillyTavern requires time and effort from you, and it rewards you with unmatched control over your experience.
Its versatility also presents a challenge for developers and contributors. Improving existing features or making major changes is complicated and often takes more effort than expected. However, that hasn’t stopped the developers from continually refining SillyTavern. They also encourage and assist contributors who want to improve the frontend.
If SillyTavern isn’t for you, there are other alternatives. But they often lack the features and customization that SillyTavern provides and might not be free or open-source. It’s difficult for new projects to match SillyTavern’s feature set, but they may be a simpler option for your needs.
