SillyTavern is a locally installed interface for talking to text generation models — and, depending on what you bolt onto it, image generation engines and text-to-speech voices as well. It is free and open source under the AGPL-3.0 licence, the project provides no hosted service of its own, and it does no programmatic user tracking. That is the entire software side of the price: nothing. What you pay for is the model backend you connect it to, and that is a separate purchase from a separate vendor.

The project began in February 2023 as a fork of TavernAI 1.2.8 and has been developed independently since. As of the research date for this guide, 9 November 2026, its GitHub repository shows roughly 35.5k stars and 6.7k forks, with more than 300 contributors.

What you are actually installing

SillyTavern is not a chat service. It is a front end: you run it on your own machine and it drives a model that runs either on the same machine, on another box on your network, or behind somebody else's API. Setup is organised around character cards — bundles of prompts that set how the model behaves, close in spirit to ChatGPT GPTs or Poe bots. Persistent conversations need a card; the docs demonstrate the concept with a default example character called Seraphina.

  • Advanced text generation settings plus community presets — sampler and prompt control is the whole point of the app.
  • World Info, the project's lorebook system for keyword-triggered context.
  • Group chats, so several characters can share a scene.
  • UI customisation through themes and custom CSS.
  • User personas.
  • A built-in RAG layer the docs call the Data Bank.
  • Slash commands and the STscript scripting engine.

Extensions are where the extra setup hides

A good part of SillyTavern's reputation comes from extensions rather than the core app: character sprites with emotional expressions, chat auto-summary, UI and chat translation, image generation through Stable Diffusion, FLUX or DALL·E (the README also notes Automatic1111 and ComfyUI integration), text-to-speech, and web search. Further extensions are available from the in-app Download Extensions & Assets menu. Each of these can mean another program to install and, for image generation, more hardware to buy.

What you need to run it

The front end itself is undemanding: anything that runs Node.js 20 or newer. That one requirement rules out Windows 7, which cannot run Node 20. The officially covered platforms are Windows, Linux and macOS, Android via Termux, and Docker.

Local model inference is a separate and optional requirement. If you intend to run the model on your own GPU, the docs recommend a 3000-series NVIDIA card with at least 6 GB of VRAM. Treat that as a rough starting point: actual needs vary with the model and the backend.

On Windows there are three documented install routes:

  • Git: install Node.js LTS and Git for Windows, clone the release branch with git clone https://github.com/SillyTavern/SillyTavern -b release, then run Start.bat.
  • SillyTavern Launcher: winget installs Git, you clone SillyTavern-Launcher and run installer.bat.
  • GitHub Desktop.

Two caveats come straight from the docs: do not install into Windows-controlled folders such as Program Files or System32, and do not run Start.bat as administrator.

There are also two branches. release is the recommended one and updates roughly monthly; staging updates several times a day and may break. For a first setup that sentence is the entire stability policy — pick release.

What it actually costs

The front end is free, with no paid tier, and the project does not accept donations for itself. Individual contributors have their own Patreon and Ko-fi pages; those are theirs, not the project's, and they buy you nothing in the app.

SillyTavern is only an interface, though, and inference costs money somewhere. The quickest no-cost start is AI Horde, which gives you out-of-the-box chatting on other people's idle capacity. Beyond that it speaks to a long list of local and cloud backends: any OpenAI-compatible API, KoboldAI, Tabby, NovelAI, OpenRouter, Claude, Mistral and others, all documented in the API Connections section of the docs.

So the real range runs from $0 — AI Horde's free tier, or fully local inference on hardware you already own — up to whatever per-token pricing your chosen cloud backend charges. Going local does not delete the cost; it moves it into the GPU from the requirements section above.

There is no official pricing page for SillyTavern itself, and every specific per-token figure you will see belongs to a backend vendor. Those numbers are out of scope for this guide, which stays on first-party sources. Our own app record for SillyTavern, filed 2026-10-27, still lists pricing as unverified for exactly that reason.

The caveats worth knowing before you start

Our app record for SillyTavern, filed 2026-10-27, holds community complaints rather than first-party ones: a steep learning curve, finicky lorebook keyword triggers, awkward mobile setup via Termux or LAN, and API bills that climb during heavy tool-use runs. That is community evidence and flagged as such — worth treating as a warning, not a verdict. We have no scam record for SillyTavern, which is an absence of reports, not a clearance.

The desk's opinion, plainly an opinion: SillyTavern is the right tool if you want full control over prompts, samplers and which model answers you, and you are willing to spend an evening on setup. It is the wrong tool if you want to sign up, pay one subscription and be chatting in two minutes. Nothing here is broken for that second person — the software simply is not built for them.