by SillyTavern (open-source community)
Free, open-source chat front end you run on your own machine and plug into any LLM, with character cards and group chats; 35,900 GitHub stars.
SillyTavern suits technical adults who want full control over a roleplay or chat setup and are willing to install software and pick a model. It is open source, keeps chats on your own machine and works with cloud or local models. Skip it if you want a phone app in a minute; the docs warn of a steep curve.
Total model freedom: the same characters and lorebooks work with a cloud API today and a local model tomorrow.
Character cards, lorebooks and extensions all work across whichever model you connect, so a user who builds a library keeps it, while the setup work keeps casual users away.
Main trade-off
Setup and upkeep that a casual user will not get through without help.
SillyTavern itself is free and open source under AGPL-3.0, read on its GitHub repository on 22 November 2026. What you pay depends on the model you connect: AI Horde is free on volunteer GPUs, OpenRouter and most cloud APIs charge per token, and a local model costs nothing per message but needs suitable hardware.
SillyTavern is a free, open-source chat interface for language models that you install on your own computer. Released under AGPL-3.0 and started in 2023 as a TavernAI fork, it connects to cloud APIs or local models and is built around character cards, lorebooks and group chats. Its docs credit more than 300 contributors.
Bring your own model
Connects to AI Horde, OpenAI-compatible APIs, Claude, Gemini, Mistral, OpenRouter, DeepSeek or local backends such as KoboldCpp, Ollama and llama.cpp.
Character cards
Each character is a card of prompts that sets how the model behaves, and community-made cards can be downloaded from inside the app.
World Info lorebooks
Store setting and lore in entries that are pulled into the prompt only when they are relevant, saving context space.
Group chats
Multi-bot rooms where several characters talk to you and to each other.
Built-in RAG
Attach documents to a chat so the model can refer to them while it writes.
Extensions
Add-ons for image generation, text-to-speech, auto-summaries of long chats, translation, web search and character expression sprites.
Scripting and commands
A chat command system and its own scripting engine for automating prompts and replies.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| Free | Yes | Android · Mac · Windows | 3.6 | |
| $9.99/mo | Yes | Web · iOS · Android | 3.1 | |
| $12.99/mo | Yes | Web · iOS · Android | 3.8 | |
| $4.99/mo | Yes | Web · iOS · Android · Mac · Windows | 4.5 |
Open source (AGPL-3.0)
The software costs nothing, so the real bill is the model behind it. OpenAI and Claude ask you to prepay credits and charge per prompt, DeepSeek needs a $2 minimum top-up, and OpenRouter is pay-per-token with some free models on limited daily requests. A model on your own computer is free per message but needs a capable graphics card.
Yes. AI Horde works out of the box on volunteer GPUs with no settings, and a local model through KoboldCpp, Ollama or llama.cpp runs on your own hardware. Wait times on AI Horde depend on the volunteers serving it.
Anything that runs Node.js 20 or higher, and install guides cover Windows, Linux and Mac, Android and Docker. For running models locally, plan on a recent NVIDIA graphics card; the docs name the 3000 series with 6 GB or more of video memory.
SillyTavern adds none of its own; behaviour depends on the backend. Its docs say local models have no content filter, while cloud APIs filter to different degrees. It has no accounts or age checks, so it is not a tool to hand to anyone under 18.
Chats, characters and settings are kept in a data folder inside the installation by default. If you run a local model, your prompts never reach an outside server. With a cloud backend, your prompts go to that provider under its own data terms.
SillyTavern is a locally installed interface for chatting with language models, image generators and text-to-speech voices. Its docs say it began in February 2023 as a fork of TavernAI 1.2.8 and now counts more than 300 contributors; the code is public on GitHub under the AGPL-3.0 licence, and version 1.19.1 shipped on 19 November 2026. It sits on the same shelf as Character.AI, Kindroid and Nomi in Hokai's list of AI chatbots, but it is a different kind of product: software, not a service.
SillyTavern holds no model of its own. You point it at a backend: a free volunteer network called AI Horde out of the box, a cloud API such as OpenRouter, OpenAI or Anthropic's Claude, or a model on your own computer through Ollama or llama.cpp. Everything else is built around character cards, a set of prompts that fixes how the model plays a character, plus lorebooks for world detail and personas that tell the character about you.
It fits hobbyists and writers who want full control over prompts and settings, people who want their chats to stay on their own hardware, and anyone moving off a hosted app who already pays for an API.
The docs are frank that it embraces a steep learning curve. You install it with Node.js, wire up a backend yourself, and get help from Discord, Reddit and GitHub issues rather than a support desk. There is no account and no age check, and what it will write depends entirely on the model you connect: the docs note that local models carry no content filter at all. If you want something that works on a phone in a minute, Character.AI or Kindroid are far easier.
So: choose SillyTavern if you are an adult who wants the most control and privacy and does not mind setup; pick a hosted companion app if you just want to start chatting.
Product facts on this page (pricing, plans, integrations, compliance) are read from SillyTavern (open-source community)'s own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.