A SillyTavern Alternative That Writes the Lorebook For You

17 September 2026

If you are looking for a SillyTavern alternative, the useful first question is which part of it you want to replace. Almost nobody dislikes SillyTavern itself. What people get tired of is the work around it: installing it, wiring up a backend, and then hand-maintaining a lorebook so the AI stops forgetting who somebody is. LLM RPG replaces the lorebook upkeep and the setup — and it gives up the model control that is the whole reason many people use SillyTavern in the first place. This article is about which of those trades you actually want.

The short verdict

• Stay with SillyTavern if you want to pick your own model, run it locally, keep your chats on your own disk, or use character cards you already have.
• Try LLM RPG if you want the world’s lore read in and indexed for you, a referee that settles what your character can actually pull off, and a game you can be playing in under a minute with nothing installed.
• You can test the second claim without deciding anything: five turns are free, with no account and no card.

Who should stay with SillyTavern

This is worth stating plainly, because a comparison that has no such section is a sales page.

SillyTavern’s own documentation describes it, as of September 2026, as “a locally installed user interface that allows you to interact with text generation LLMs, image generation engines, and TTS voice models”, whose goal is “to empower users with as much utility and control over their LLM prompts as possible, embracing the steep learning curve as part of the fun.” That sentence is not a warning label — it is a description of the audience, and if it describes you, no hosted product is going to beat it. Specifically, keep SillyTavern if any of these matter:

• You want to choose the model. SillyTavern connects to a long list of local and cloud backends. LLM RPG does not let you pick one at all.
• You want everything local. Chats, characters and lorebooks live in your own installation directory. LLM RPG is a hosted service.
• You use character cards, image generation or TTS. LLM RPG has none of these — it is one text RPG, not a general chat frontend.
• You want free and open source. SillyTavern is AGPL-3.0 and its docs promise it “will always be free and open-sourced”. LLM RPG is a paid product with a free tier.

Where the work actually goes in SillyTavern

Two things have to be true before your first message, and both are on you.

The model is not included. SillyTavern’s docs are explicit: “Since SillyTavern is only an interface, you will need access to an LLM backend to provide inference.” AI Horde is offered for instant out-of-the-box chatting, and beyond that it is an OpenAI-compatible API, KoboldAI, Tabby and many more. If you want to run inference on your own machine, the docs recommend “a 3000-series NVIDIA graphics card with at least 6GB of VRAM” — the interface itself will run on anything that can run Node.js 20.

The world is not included either. Lore in SillyTavern lives in World Info, also called Lorebooks, and the docs describe the mechanism precisely:

“It functions like a dynamic dictionary that only inserts relevant information from World Info entries when keywords associated with the entries are present in the message text.”

That is a keyword trigger, written by you, per entry. The same page advises that “each World Info entry should have a comprehensive, standalone description” while also keeping entries concise to conserve prompt tokens — and it is candid about the ceiling:

“It is important to note that while World Info helps guide the AI toward the desired content, it does not guarantee its appearance in the generated output messages.”

For a small scenario that is fine, and the control is genuinely nice. For a novel with sixty named characters, four factions and a map, it is a second hobby: you are the librarian, and every gap in your keywords is a moment where the AI confidently invents something.

How LLM RPG handles the same problem

LLM RPG starts from the other end. You give it the book — a .txt or an .epub — and before you play, it reads the whole thing and builds a searchable atlas of the characters, factions, locations, items and events in it. During play, the relevant parts of that atlas are recalled by meaning rather than by keyword matching, so a scene about someone’s brother surfaces what the book says about the brother whether or not you thought to list his name as a trigger. When the book later contradicts an earlier fact, the superseded version is retired rather than left to compete as canon.

If you have no book to hand, there are 24 ready-made worlds — fantasy, wuxia, cyberpunk, horror and others — each with its lore already built, so a run starts in seconds.

The part that is not a chat at all

The second difference is bigger than the lorebook, and it is the reason this is filed as a game rather than as a roleplay frontend.

In a chat frontend, the model narrates and decides. You say your character picks the lock; the same model that is telling the story decides whether the lock opens, which means outcomes drift with the prose and cannot be learned.

In LLM RPG, those two jobs are split. A referee reads your declared action against your character sheet and against the lore recalled for that scene, and reports facts — what ability this draws on, how demanding the moment is. The narration model never produces the number and never sees the roll. The result is settled against odds, and only then written up as a scene. It means a long-shot can fail, and that failing is part of the game rather than the AI being in a bad mood.

Side by side

What each one costs

SillyTavern is free. Your real bill is inference: either an API backend you pay per token, or a GPU you already own. That bill is invisible in a comparison table and very visible in a monthly statement, so it is worth being honest that “free” here means “free software”, not “free to play”.

LLM RPG is metered in credits, and the whole price list is public:

One credit is one narrated turn on Free and Pro. Reading a book into lore costs 1 credit per 50,000 characters, and you are shown the exact price of a book before anything is spent. Failed turns and unfinished extractions are refunded automatically. The full breakdown is on the pricing page.

How to switch, if you want to

There is no import path, and pretending otherwise would waste your evening: character cards, chat logs and World Info files do not carry across. The migration is really a re-start, and it is short.

1. Open the play page and take five free turns in a ready-made world. No account, no card. This tells you within about ten minutes whether the referee model of play is one you enjoy.
2. If it is, upload the novel you were building a lorebook for. Chapters and language are detected for you; you are shown what the extraction will cost before it runs.
3. Create your character and play. What you were maintaining by hand is now the thing being queried automatically.

Keep your SillyTavern install. They are not the same tool, and the trade is real in both directions: you are exchanging model control and local ownership for not having to be your own librarian. If the lorebook upkeep was the part you dreaded and the model tinkering was the part you enjoyed, this is the wrong trade for you — and the five free turns will tell you that faster than another comparison article will.

Claims about SillyTavern above are quoted from its own documentation and homepage as of September 2026; check them there, since it ships often — its stable branch is updated roughly monthly and its staging branch several times a day.