WyvernChat vs SillyTavern on mobile

If you read my hiatus post, then you know that I’ve been a lot more busy than I’d like to be lately. As a result, I’ve been mostly sticking to RP on mobile. This isn’t a full WyvernChat vs SillyTavern breakdown, but it explains why I’ve been reaching for WyvernChat more lately. I’m a big SillyTavern guy. I first started on CrushOn, then SpicyChat, then Chub.ai, and then I started looking for a bit more control. That’s when I came across SillyTavern on a reddit post, which was expressing shock that people unironically use Chub.ai (I was one of them! Paying for that Mars subscription!) and not just using it as an easy place to get character cards. This shocked me! I didn’t even know what SillyTavern was, never heard of it.

Skip forward to today, and I still run an instance off a Pi 4 and use OpenRouter as my main way to get ahold of an LLM. I use Tailscale to connect to it from anywhere, and that’s perfectly fine. However, SillyTavern’s UI on mobile is painful (maybe it’s just me..) and I hated it. It’s a great open-source project with a lot of love behind it, and I could never hate it in that way. It just leaves a little bit desired. So I started looking into an alternative. That’s when I came across WyvernChat, and man; I just fell in love. The biggest thing for me was how little friction there was between opening the site and actually getting into a session.

my first impressions

The UI has a few kinks here and there, but man is it smooth. The character library is already THERE, so no cloud importing or none of that, unless you want a character that isn’t on the platform. Even then, it’s seamless. Much like Chub, you can choose to archive or publish it. Speaking of the character library, they have a verified creator column with some great character cards that I’ve enjoyed. In general, the character library is pretty high quality.

That probably sounds like a small thing, but coming from SillyTavern, it changes the whole experience on mobile. SillyTavern gives you an absurd amount of control, but that control also means there are more steps. WyvernChat feels much closer to opening an app, picking a character, choosing a model, and just going. That convenience would mean a lot less if WyvernChat stripped away all the control that made me move to SillyTavern in the first place, though. Thankfully, it really doesn’t.

However…

SillyTavern is still on another level when it comes to customization. You can tweak just about everything if you’re willing to dig through the settings, extensions, presets, and prompt formatting. WyvernChat isn’t trying to completely replace that. What surprised me is how much of the control I actually care about is still there, just presented in a much cleaner way. I’ve realized that I don’t always need all of that control. Most of the time, I want enough control to make the model behave the way I want, enough flexibility to use my own characters and prompts, and then I want to start roleplaying. WyvernChat gets really close to that sweet spot.

And once you add its model options into the equation, that’s where things get even more interesting. WyvernChat has an incredibly generous free tier, but the skies the limit, specifically with the Featherless subscription (though it does support OpenRouter).

The Featherless Subscription

If you’ve been reading the blog for a while, you’ll know that I used to be a huge z.AI fan. Coding Lite used to be like, $7 and you used to be able to RP with GLM infinitely. There was no token limit, and it was pretty snappy too. z.AI got popular, and these things get expensive over time, so they changed their pricing/usage models. I love OpenRouter, but I like subscription-services that favor unlimited use. I don’t really like usage anxiety or worrying about if I’m gonna hit a limit or not. Nor are I a big fan of seeing myself losing 0.01c every time I send a prompt. I’d rather just pay a lump sum every month and not have to worry about it. That’s where the Featherless subscription comes in.

Featherless’s Subscription Model

Featherless is $25 a month. In exchange you get unlimited use across practically ANY Huggingface model (which, at the time of this article, is 45k+). These include fine-tunes, and the latest major models you may know that are open-weights. (At the time of this article, being GLM 5.2, Kimi K3, DeepSeek V4 Flash/Pro, MiniMax, Qwen 3.8-27B, Gemma 4 31B, just to name a few!) Sounds great, right? It does come with one important caveat that almost completely steered it away for me:

You’re limited to 32k worth of context.

When I first read that, I thought it was practically unusable. A good preset, a character card between 1-2k tokens, lorebooks, etc, eats a decent chunk of that. You’re down to about 26k worth of rolling context. I decided “Well, what the hell, it’s $25, I spend more on stupid crap all the time. Worst case scenario, it’s $25 down the drain”. So I got it. And let me tell you, it did not disappoint.

WyvernChat’s Memory and Context Features

WyvernChat, natively, includes several tools that facilitate stretching your tokens and context as much as possible.

Summarization

It has built-in chat summarization with a custom prompt (here’s mine!)

Summarization Prompt

Summarize the conversation above in approximately 250–350 words. The summary will be used to restore context in later roleplay messages, so prioritize continuity and durable memory over literary style. Preserve: * Significant events, actions, and outcomes * Important decisions, promises, goals, plans, and instructions * Character development, emotional shifts, and relationship changes * New or changed information about characters, including motivations, knowledge, abilities, injuries, possessions, and ongoing conditions * Relevant worldbuilding, locations, factions, rules, objects, and established lore * Unresolved conflicts, mysteries, obligations, and immediate next steps * Information individual characters learned, when that distinction matters Prioritize recent developments while retaining older details still necessary to understand them. Clearly distinguish confirmed facts from suspicions, misunderstandings, lies, and assumptions. Avoid repeating the same information in both an overview and later sections. Omit minor banter, disposable scene description, and static character details unless they became relevant during the conversation. Do not invent, infer, embellish, analyze, or continue the story. Do not treat speculation as fact. Write a dense, factual summary using clear paragraphs or compact bullets. Reserve the final paragraph for the exact current scene, including location, present characters, physical conditions, emotional state, and the immediate unfinished action. Always finish the current-scene paragraph, even if earlier details must be shortened.

This is pretty self explanatory if you’ve been in the space for any length of time, but essentially it’s summarizing parts of your story in order to shorten the amount of tokens you use. It’s useful to both lower costs and also to keep context tight, helping the AI focus on more details. Summarization by itself isn’t anything revolutionary. I’ve been summarizing long chats for ages. Where WyvernChat started to impress me was how many of these tools are built directly into the platform and how they work together.

Memory Scan

WyvernChat also has a Memory Scan tool. Quoting the platform directly, “Memory Scan reads your chat and proposes lexicon entries — memories, NPCs, locations, events — so the AI remembers them later. Review and keep what you want.” You can choose what model to use, how many messages to scan per request, and what specific model/preset you want to use in order to conduct this scan. That means you can use an excellent summarization model (I use Gemma 4 31B, for example) and pair it with a specific system prompt/final instructions to really extract as much information from your chat as possible, and turn them into categorized lorebook entries. It is super intuitive, integrated, and easy to use.

Sidechat

It includes a side chat feature where you can discuss your role-play with an AI of your choosing with a custom configuration (which is essentially an AI with context of your story, in case you want to brainstorm or garner ideas). I’ve found this surprisingly useful when an RP gets complicated. Instead of asking the actual roleplay model an OOC question and muddying the conversation, I can jump into Side Chat and ask something like, “What unresolved plot threads do I currently have?” or “Would this character realistically know about X?” without touching the actual session.

The Lexicon

The Lexicon is also a feature, which is also an AI that you specifically articulate which parts of the stories you want saved or updated into a lexicon entry. You can ask the assistant to document lore from your session, expand more, create new characters, or other story elements from the convenience of a separate conversation that’s already integrated into your RP. It becomes brainstorm -> create -> implement. A flawless loop, especially convenient on a mobile device.

So is 32K worth of Context Enough?

If you decide to go with Featherless!

On paper, no, not really. By itself, 32k context is astronomically small compared to current LLM capacity. Now-a-days, any model that doesn’t support up to 1M tokens in context is considered less capable. The reality is, for RP purposes, context matters less than you think, especially with the comprehensive tools that WyvernChat offers. By itself, none of this is revolutionary, and you don’t have to struggle to fit your story into 32k tokens because WyvernChat supports OpenRouter and other services perfectly fine. They have their own context limits depending on which model/service you use.

But, honestly, I’ve also started questioning how useful gigantic raw context windows actually are for RP. I’d rather give the model 25K tokens of relevant recent conversation plus a good summary and curated memory than dump an entire novel into the context and pray it pays attention to the right parts. I’ve been using ChatGPT to make a “bible” of sorts, which is essentially a very curated and detailed summary that’ll eat about 4-5k tokens. I’ve managed to make one of my most favorite sessions last 800+ messages this way (still going strong, too!). Then, I update the bible every arc or so, which only takes a second thanks to ChatGPT’s massive context window and 5.6 just being pretty damn good.

So, am I replacing SillyTavern with WyvernChat?

So, am I moving everything over to WyvernChat? Absolutely not. Also, I’m not a shill, I swear. No affiliate marketing here, bucko.

SillyTavern is still SillyTavern. If I want to sit at my computer and obsess over prompts, extensions, formatting, lorebooks, or whatever weird configuration I’ve convinced myself is going to finally create sentient life, that’s where I’m probably going.

But that’s not how I’ve been RPing lately.

Most of the time I’m on my phone, I want to open something, pick my character, pick my model, and go. WyvernChat gives me that without making me feel like I’ve traded away all the control I went to SillyTavern for in the first place. Add Featherless, Memory Scan, summaries, and the Lexicon, and I’ve somehow managed to keep extremely long stories going inside a context window that I originally thought was laughably small.

That’s where I land on WyvernChat vs SillyTavern. WyvernChat isn’t objectively better at everything. It just fits the way I’m actually using AI right now.